Kimi models compared cost vs quality at every effort level

Data analysis · aggregated from 87 sources · 444 benchmarks · 576 measurements
GitHub
Alexandre Gensse

What each recent Kimi model actually costs, at every effort level, and the quality it delivers — fused from public benchmark measurements for every (model, effort) pair, on the same scales as every other maker's on this site.

The best value today is Kimi K2.7 Code — 1.1× cheaper than the price trend for its quality (expected score 54.9 % on the benchmark panel).

Quality is the expected score across the whole benchmark panel, not one benchmark: the type of task can move it, see what moves everything.

Measurements
87 sources · 444 benchmarks · 576 measurements
Updated
19 Nov 2026
👑 Best overall
Kimi K2.7 Code Cost 2.9× · Score 54.9 %
114
Grunt work
Kimi K2.6 · Thinking off Cost 1.0× · Score 43.6 %
55
Classification, tagging, extraction, routing, log/PR triage — run at scale, where throughput and unit cost dominate.
Everyday tasks
Kimi K2.7 Code Cost 2.9× · Score 54.9 %
114
Routine coding, refactors, unit tests, summaries, first-draft agent steps — solid work that doesn't need the frontier.
Advanced reasoning
Kimi K3 · Max Cost 9.6× · Score 62.1 %
99
Production code review, architecture, hard debugging, customer-facing reasoning — you need essentially flagship quality.
Cutting-Edge thinking
Kimi K3 · Max Cost 9.6× · Score 62.1 %
99
Research-grade reasoning, novel or ambiguous problems, the hardest agentic runs — a few extra points of capability are worth a premium.

Quality vs. cost of every Kimi model, at each effort level

One point per couple: a model at one configuration (thinking off, minimal, low → max, thinking on); no line joins a model's couples. Only the couples of the Pareto frontier are labelled (model · effort), where there is room; hover or tap any point for its model, effort, cost and score. Cost is a multiple of the cheapest couple shown; quality is placed on the model's latent scale and labelled with the score it predicts on the benchmark panel. Below the weakest tier target the quality axis is compressed, so the models short of today's level sit together at the bottom. Tier bands (off by default) shade the four usage tiers of the tier picker below : each band runs from that tier's target up to the next tier's — the qualities that clear its bar but not the next one. Uncertainty ovals (off by default) show each couple's 16–84 % interval, width = cost, height = quality; hover to bring the one under the cursor forward.

Best Kimi model for the price — the Pareto frontier

The non-dominated (model, effort) couples : nothing offers more quality at equal-or-lower cost. Solid = frontier, faded = dominated; the frontier couples are labelled (model · effort) where there is room — hover or tap any point for its model, effort, cost and score. No line joins a model's couples. The frontier couples are the candidates of every pick below. The faint grey line is the price trend — what a given quality costs today, fitted on every couple of the latest model of each family. Tier bands (off by default) shade the same four usage tiers as on the landscape above.

Value — distance to the price trend

The price trend — ln(cost) = a + λ·θ, θ the latent quality — is what a given quality costs today across the 4 couples of the latest model of each family, not just the frontier: it is the going rate of the models on sale. Earlier generations, dearer per point than their successors, would flatten it and make quality look cheap; it does not move when older models are shown (R² ≈ 0.97, drawn faint grey above). Each couple weighs by its uncertainty across the line, both axes counted. On the latent scale the trend is a straight line: each step of quality costs the same ratio more, wherever it sits. A couple's distance to it says how many times cheaper or dearer it is than the trend charges for its quality — the same gap read on the other axis is how much more quality it gives than the trend at its cost. No couple serves as a reference. Only the frontier couples are listed.

Model · effortcostexpected scorevalue index
trend = 100
Kimi K2.7 Code · solo2.9×54.9 %114
Kimi K3 · Max9.6×62.1 %99
Kimi K3 · High8.6×60.2 %84
Kimi K3 · Low2.6×51.8 %81
Kimi K2.6 · Thinking off1.0×43.6 %55

Value index = 100 × how many times cheaper than the trend the couple is for its quality (100 = the trend). Green = above 100, red = below; intensity = distance to the trend in decades, capped at one. A couple on the Pareto frontier can still be red : being non-dominated does not mean being good value for what it delivers.

Best value by task tier — what to actually run

Consolidating the frontier into a decision. Four target qualities are derived from the data, not fixed: they spread evenly between the frontier's weakest and strongest couples, each end drawn 5 % of the span inward, so that the bottom tier follows the weakest couple available and the top tier the best — as models improve, the targets move with them. The width follows the spacing on one rule : adjacent windows cross at half weight midway between their targets. For each tier we pick, among the frontier couples, the one that maximises window × eβ·λθ ⁄ cost, λ the slope of the price trend and β the tier's price of quality as a multiple of the market's : grunt work pays a quarter of it (β = 0.25), everyday tasks 0.63, advanced reasoning 1.59, cutting-edge thinking four times (β = 4) — evenly spread in ratio, symmetric about the market. Below its target a couple is penalised by a Gaussian window ; at or above it the window is 1, and eβ·λθ ⁄ cost rewards smoothly a couple that brings more quality than the tier is ready to pay for its extra cost. Cost counts in ratios, as people perceive prices : twice as dear weighs the same at every price. The big figure is the pick's value index : 100 = the price trend, 500 means five times cheaper than the trend charges for its quality, 50 twice as dear. The crown is the frontier couple that sits furthest below the trend. Sliders below override the targets, widths and prices of quality.

Target 44.4 % – Grunt work
Kimi K2.6 · Thinking off Cost 1.0× · Score 43.6 %
55
Classification, tagging, extraction, routing, log/PR triage — run at scale, where throughput and unit cost dominate.
Target 49.6 % – Everyday tasks
Kimi K2.7 Code Cost 2.9× · Score 54.9 %
114
Routine coding, refactors, unit tests, summaries, first-draft agent steps — solid work that doesn't need the frontier.
Target 55.2 % – Advanced reasoning
Kimi K3 · Max Cost 9.6× · Score 62.1 %
99
Production code review, architecture, hard debugging, customer-facing reasoning — you need essentially flagship quality.
Target 61.2 % – Cutting-Edge thinking
Kimi K3 · Max Cost 9.6× · Score 62.1 %
99
Research-grade reasoning, novel or ambiguous problems, the hardest agentic runs — a few extra points of capability are worth a premium.
👑 Best overall
Kimi K2.7 Code
Cost 2.9× · Score 54.9 % · Value index 114

Picked as the frontier couple that sits furthest below the price trend : it costs 1.1× less than the trend charges for its quality — or, read on the other axis, it scores 0.9 points of expected score above what the trend gives for its cost. Its value index is that ratio times 100: 100 = the trend, fitted on every couple of the latest model of each family, so today's going rate of the models, not the frontier; no couple serves as a reference.

Tune the tiers (target, σ, price of quality β)
Normalized cost matrix — Kimi effort levels compared

Cost per model × effort as a multiple of the cheapest couple, sorted by decreasing expected score. Every cell is computed from the measurements by the fusion model described in the method below: a cost on a task of typical size, with the couple's own 16–84 % interval. Intensity = cost.

Modelthinking offlowmediumhighxhighmaxthinking onExpected score
top effort
Kimi K3—
2.62.2–3.0
—
8.65.0–15
—
9.68.6–11
—62.1 %
Kimi K2.7 Code
2.9merged · 2.4–3.6
54.9 %
Kimi K2.6
1.00.81–1.2
—————
3.92.8–5.5
53.8 %

"Expected score · top effort" = the score the model at its top effort is expected to reach on the benchmark panel, from the same fit as the landscape. A model measured at a single setting (solo : no effort parameter, see the measurement sources) fills one merged cell. A thinking off column appears when a model is measured with its thinking switched off, a minimal one for the lowest level some models have below low; each is a configuration of its own, read from its own measurements. n/a : effort not exposed, or measured by a single publisher (a couple is shown only once two publishers measured it).

Value-index matrix — every model and effort against the price trend

The value index of every couple shown, model × effort, in the same order as the cost matrix : 100 = the price trend, what a given quality costs today across the latest model of each family ; 500 means five times cheaper than the trend charges for that quality, 50 twice as dear. Under each value, its 16–84 % range.

Modelthinking offlowmediumhighxhighmaxthinking on
Kimi K3—
8153–122
—
8447–152
—
9985–116
—
Kimi K2.7 Code
11488–148
Kimi K2.6
5540–77
—————
7450–108

Green = above 100, cheaper than the trend for its quality ; red = below ; the deepest green is the best index shown, the deepest red the lowest, on a log scale. A high index does not make a couple a pick : the tiers weigh it against their target quality.

What moves everything — four factors behind the numbers

The tier picks above are an order of magnitude. Four factors, each measured elsewhere, modulate them strongly — keep them in mind when reading the tiers and the matrix.

Task type

Effort and model choice don't weigh the same depending on the nature of the work — review, reasoning and pure coding reward effort very differently.

Complexity

The harder the task, the more effort costs — up to the point where extra thinking starts to hurt (the inverted-U).

Generation

Each release buys the same quality for less: turn on Older models to set each family's earlier models beside its current one, on the same scales.

Cache & harness

Two invisible factors — cache-read rate and the surrounding agent harness — that can dominate both cost and score.

Data Sources

Every source below measured at least two (model, effort) couples on the same task; 10115 cost and 23548 score measurements enter the fusion. Each row below is one such source, with the actual configuration we verified (harness · effort), its type, and the couples it links.

SourceActual config (harness · effort)TypeLinked couples (same task)
AA CritPt
artificialanalysis
AA model payload · CritPt physics, 4 models × ladder + 3 at max; floor scores blank ✓sweepKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max · Kimi K2.6·nothink
AA GDP-PDF
artificialanalysis
AA model payload · GDP-PDF all-pass, 4 models × ladder + 3 at max; noisy ✓sweepKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max
AA GDPval v2
artificialanalysis
AA evaluation page · 220 tasks, Elo; cost derived from AA's published token counts ✓sweepKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max
AA Harvey LAB
artificialanalysis
AA model payload · Harvey's Legal Agent Bench (AA's own run, separate from Vals's HLAB), score-only, 8 models × ladder + 4 at max ✓sweepKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max
AA HLE
artificialanalysis
AA model payload · HLE no tools, 4 models × ladder + 3 at max ✓sweepKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max · Kimi K2.6·nothink
AA Index v4
artificialanalysis
AA model pages · Fable 5.1 sweep low→max + Fable 5/Opus 5 at max, per-suite $ ✓sweepKimi K2 Thinking·solo
AA MLCR
artificialanalysis
AA model payload · MLCR-AA, score-only; Sonnet 5.5 max-only, Opus 5 full ladder, 5 others at max ✓sweepKimi K2.7 Code·solo · Kimi K3·max
AA Omniscience
artificialanalysis
AA model payload · Omniscience accuracy (not the penalised index) ✓sweepKimi K2.6·think · Kimi K2.6·nothink · Kimi K2.7 Code·solo · Kimi K3·max
AA SciCode
artificialanalysis
AA model payload · SciCode, 4 models × ladder + 3 at max ✓sweepKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max
AA TB 4.0
artificialanalysis
AA model payload · Terminal-Bench 4.0 (AA harness), 4 models × ladder + 3 at max; floor scores blank ✓sweepKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max
AA-Briefcase
artificialanalysis
AA model payload 27 Sep · agentic knowledge work, Elo; 4 models × full ladder + 3 at max; costPerTask per eval ✓sweepKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max
AA-LCR
artificialanalysis
AA model payload · long-context reasoning, input-dominated cost ✓sweepKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max · Kimi K2.6·nothink
Agents on Rails · features
rubyonrails
rubyonrails.org/ai + rails/ai-evals runs · 20 tickets ×3, miniswen; Opus 5 & Fable 5.1 high→max, Opus 5.5 default ✓sweepKimi K3·max
ARC-AGI-1
arcprize
ARC Prize evaluations.json · semi-private, costPerTask per model × effort; scores ≥97.5 % flagged gen5-saturated ✓sweepKimi K3·max
ARC-AGI-2
arcprize
ARC Prize evaluations.json · semi-private, costPerTask per model × effort; Opus 5.5 & Fable 5.1 full ladders ✓sweepKimi K3·max
BullshitBench v2
github-bullshitbench
petergpt/bullshit-benchmark c1d4ab6 · 100 nonsense prompts, share of clear push-backs (green), effort from the board's reasoning column; primary leaderboard ✓sweepKimi K2.6·nothink
CursorBench 3.2
cursorbench
Cursor 3.2 · matched effort, $ cost ✓ (vendor benchmark)sweepKimi K3·high · Kimi K3·low · Kimi K3·max
InfluMatch Set 1 (11 queries) P@5, reason on all 50 candidates
arxiv-2607.05968
KOL search P@5 %, Kimi-K2.6 as reason scorer think vs no-think; tokens per query from Table 10 (no USD); paper 2026-07-07; read 2026-10-20 ✓sweepKimi K2.6·think · Kimi K2.6·nothink
InfluMatch Set 1 P@5, SimPO rerank then reason top-10
arxiv-2607.05968
KOL search P@5 %, Kimi-K2.6 as reason scorer think vs no-think; tokens per query from Table 10 (no USD); paper 2026-07-07; read 2026-10-20 ✓sweepKimi K2.6·think · Kimi K2.6·nothink
InfluMatch Set 2 (31 queries) P@5, reason top-10 after SimPO rerank
arxiv-2607.05968
KOL search P@5 %, Kimi-K2.6 as reason scorer think vs no-think; tokens per query from Table 10 (no USD); paper 2026-07-07; read 2026-10-20 ✓sweepKimi K2.6·think · Kimi K2.6·nothink
ObviousBench 0.2
benchmarklist
inspect-ai · 144 items ×3 epochs, 3 models × full ladder, measured tokens ✓sweepKimi K2 Thinking·solo · Kimi K2.7 Code·solo
Playcode MacBook SVG
playcode
Playcode · one-shot MacBook SVG, high/xhigh/max per model, provider-billed cost (quality ordinal, not scored) ✓sweepKimi K3·high · Kimi K3·max
PostTrainB.
arxiv-2603.08640
papier · medium/high ✓sweepKimi K2 Thinking·solo
Vals VibeCode
vals.ai
Vals AI · Opus 5 sweep low→max (scores), $ cost at leaderboard setting; board entries — snapshot 27 Sep, re-priced ✓sweepKimi K2.7 Code·solo
vlm-exam
github-vlmexam
roboflow · identical 133-image set, 3 models × low/high, measured tokens ✓sweepKimi K2.6·nothink · Kimi K3·high
WeirdML
epoch-archive
Epoch archive · 3 models × high/max, cost per run ✓sweepKimi K3·max · Kimi K2.7 Code·solo
Zapier AutomationB.
benchmarklist
AutomationBench 1.0.6 · 657 held-out tasks, strict pass/fail, measured tokens × list ✓sweepKimi K2.7 Code·solo
deepseek-v41-techreport: deepswe
deepseek
DeepSWE v1.1 (Resolved), mini-SWE harness; DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness mini-SWE; all columns at max effort for report/card E; 1017 card effort not stated; + effort sweep 25..100 of deepseek-v4.1-flash from Fig 9/11 (same runs; effort-100 point = Table 3 cell, kept once), tokens_out per response/trajectory; read 2026-10-19 ✓sweep+xmodelKimi K3·max
deepseek-v41-techreport: gpqa
deepseek
GPQA Diamond (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness none(reasoning); all columns at max effort for report/card E; 1017 card effort not stated; + effort sweep 25..100 of deepseek-v4.1-flash from Fig 9/11 (same runs; effort-100 point = Table 3 cell, kept once), tokens_out per response/trajectory; read 2026-10-19 ✓sweep+xmodelKimi K3·max
deepseek-v41-techreport: mathapex
deepseek
MathArena Apex (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness none(reasoning); all columns at max effort for report/card E; 1017 card effort not stated; + effort sweep 25..100 of deepseek-v4.1-flash from Fig 9/11 (same runs; effort-100 point = Table 3 cell, kept once), tokens_out per response/trajectory; read 2026-10-19 ✓sweep+xmodelKimi K3·max
deepseek-v41-techreport: tb21
deepseek
Terminal-Bench 2.1 (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness DeepSeek-Harness-minimal; all columns at max effort for report/card E; 1017 card effort not stated; + effort sweep 25..100 of deepseek-v4.1-flash from Fig 9/11 (same runs; effort-100 point = Table 3 cell, kept once), tokens_out per response/trajectory; read 2026-10-19 ✓sweep+xmodelKimi K3·max
FeatherBench
ed-is-ai/featherbench
OpenRouter(Anthropic) · 28-task multi-domain suite (coding/data/realworld/security/tool-use), effort=high (Haiku=solo), mean cost/tokens per task-run, consolidated results/summary.json; only Opus 5.5 rung saturated ✓xgenKimi K3·max
Official tau2-bench leaderboard, banking_knowledge (AllTools), Claude submissions by Sierra
github-sierra-tau2-bench
banking_knowledge domain, 4 trials, user simulator gpt-5.2 (low), reasoning_effort max for all; pass^1 %; read 2026-10-18 ✓xgenKimi K3·max
AA AutomationB.
artificialanalysis
AA evaluation page · 657 tasks partial-credit, 3 models at max ✓xmodelKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max
AA FrontierCode 1.1 (AA-run, weighted score)
artificialanalysis
Artificial Analysis AA FrontierCode 1.1 (AA-run, weighted score); score only from the model-page payload; effort from payload effort.slug; read 2026-10-15 ✓xmodelKimi K3·max
AA MMMU-Pro
artificialanalysis
Artificial Analysis AA MMMU-Pro; score only from the model-page payload; effort from payload effort.slug; read 2026-10-15 ✓xmodelKimi K2.6·think · Kimi K3·max
AA-AnalystAgent
artificialanalysis
Artificial Analysis AA-AnalystAgent; score only from the model-page payload; effort from payload effort.slug; read 2026-10-15 ✓xmodelKimi K3·max
aa-index411
tabbit.ai
—xmodelKimi K3·max
aaapex
artificialanalysis
—xmodelKimi K3·max · Kimi K2.6·think
aaenterpriseops
artificialanalysis
—xmodelKimi K2.6·think · Kimi K2.7 Code·solo
aagpqa
artificialanalysis
—xmodelKimi K3·max · Kimi K2.6·think · Kimi K2.6·nothink · Kimi K2.7 Code·solo
aaifbench
artificialanalysis
—xmodelKimi K2.6·think · Kimi K2.6·nothink · Kimi K2.7 Code·solo
aaomni-index
artificialanalysis
—xmodelKimi K2.6·think · Kimi K2.6·nothink · Kimi K2.7 Code·solo · Kimi K3·max
aatau-banking
artificialanalysis
—xmodelKimi K2.6·think · Kimi K2.7 Code·solo · Kimi K3·max
aatau2
artificialanalysis
—xmodelKimi K2.6·think · Kimi K2.6·nothink · Kimi K2.7 Code·solo
aatb21
artificialanalysis
—xmodelKimi K3·max · Kimi K2.6·think · Kimi K2.7 Code·solo
aatbhard
artificialanalysis
—xmodelKimi K2.6·think · Kimi K2.6·nothink · Kimi K2.7 Code·solo
Agent Island social-deduction arena (Plackett-Luce skill), OpenRouter models
arxiv-2605.04312
Agent Island (arXiv 2605.04312, 5 May 2026): 999 multi-agent social-deduction games with 49 models via OpenRouter, Bayesian Plackett-Luce posterior skill (mean of lambda; prior mean 1; Table 1); reasoning effort not stated; no cost; read 2026-10-20 ✓xmodelKimi K2 Thinking·solo
Agents on Rails · atomic
rubyonrails
rubyonrails.org/ai + rails/ai-evals runs · 21 Writebook tasks ×3, miniswen, list-price cost/run ✓xmodelKimi K3·max
aibenchy.com private benchmark suite — Haiku 5.5 (xhigh) vs Opus 5.5 (high)
aibenchy
aibenchy.com undisclosed-task private suite, score 1-10 scale (site's own verdict text quotes '9.1 vs 9.4'); page states 'Last updated at: 2026-10-07'; Haiku-5.5-xhigh row is the same run repeated across two /compare/ pages (identical score/cost/time/rank), not two independent measurements; read 2026-10-08 ✓xmodelKimi K2.6·nothink
AIME 2025 (GLM-4.7 card)
zhipu
AIME 2025; GLM-4.7 card 2026-01-29; settings: default settings; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
ALE-Bench
epoch-archive
Epoch archive · 3 models at matched high, measured token splits ✓xmodelKimi K3·max · Kimi K2.7 Code·solo
android-bench-2
google
—xmodelKimi K3·max
APEX-Harm mean task score under attack (all 6 categories)
apexharm-2610.13623
APEX-Harm 180 attack task instances per model (15 underlying APEX-Agents tasks x 12 attack variants), APEX-Agents harness with MCP tools, 150 steps, no timer, judge/rules per paper; metric mean task score 0-1 (utility under attack); ASR/ER safety rates not extracted; arXiv 2610.13623 Table 4 (Oct 2026); cost = Table 1 total for Gemini 3.8 Flash only; read 2026-10-20 ✓xmodelKimi K3·max
apexharm-asr
apexharm-2610.13623
—xmodelKimi K3·max
arxiv-2610.13075-longbench-v2-direct
arxiv-2610.13075
—xmodelKimi K3·max
arxiv-2610.13075-officeqa-v1-direct
arxiv-2610.13075
—xmodelKimi K3·max
arxiv-2610.13075-officeqa-v2-direct
arxiv-2610.13075
—xmodelKimi K3·max
arxiv-2610.22599-direction
arxiv-2610.22599
—xmodelKimi K3·max
arxiv-2610.22599-r2
arxiv-2610.22599
—xmodelKimi K3·max
arxiv-2610.22599-slope
arxiv-2610.22599
—xmodelKimi K3·max
arxiv-2610.22599-top-idea-gain
arxiv-2610.22599
—xmodelKimi K3·max
CL-Bench
clbench.com
CL-Bench, learning from context · overall score on 0–1 with its std in the flags; Opus 4.5 with and without thinking as default/nothink; no cost; read 2026-09-30 ✓xmodelKimi K2 Thinking·solo
ClawProBench open dataset 102 tasks (OpenClaw)
clawprobench-board
ClawProBench (suyoumo.github.io/bench), open dataset 102 tasks, OpenClaw agent runtime v3.2.6, deterministic grading, repeated trials; Final Score 0-100 (page default rank signal); cost = page cost_usd = (in + 0.5 x (cacheread + cachewrite)) x price_in + out x price_out; rows updated 2026-02..2026-07 (live board); one route kept per couple (earliest updated); read 2026-10-20 ✓xmodelKimi K2 Thinking·solo
cliphou.se Motion Graphics Benchmark v1.1: cost per run (Remotion videos)
cliphou.se
3 briefs x 2 runs via OpenRouter, reasoning high; cost = mean OpenRouter generation cost per run over the 6 runs (computed by us); no quality score published (human review pending); read 2026-10-15 ✓xmodelKimi K3·high
Conceptual Reasoning Index · DTBench
conceptualreasoning.ai
Conceptual Reasoning Index DTBench sub-score (CRI-rescaled accuracy, x100), cost = average USD per response (site cost chart); effort from the csv effort column; blank = default; read 2026-10-15 ✓xmodelKimi K3·max
Conceptual Reasoning Index · LMCA
conceptualreasoning.ai
Conceptual Reasoning Index LMCA sub-score (Pearson-r based, x100), cost = average USD per response (site cost chart); effort from the csv effort column (max rows carry the base name); blank = default; read 2026-10-15 ✓xmodelKimi K3·max
ctala
github-ctala
Claude Code CLI · reasoning not configured, temp 0.7 ✓xmodelKimi K2.7 Code·solo · Kimi K2 Thinking·solo
DataLLM Lab: 9 executed Python tasks, Qwen3.8 Max family (Prime, Max, Max-0902), pass and cost per 1,000 tasks
datallmlab
9 executed Python tasks (python3 -I subprocess, 12 s timeout), temperature 0, max_tokens 4000, one scored attempt, via OpenRouter; three runs on different days (Prime 2026-10-02, Max 2026-08-20, Max-0902 2026-09-16); cost = measured tokens x list price on the priced-on date (source-computed, 'not a billing statement'); read 2026-10-20 ✓xmodelKimi K3·max
datallmlab-coding9
datallmlab
—xmodelKimi K2.7 Code·solo
DeepSeek-V3.2 report: aime
deepseek
aime; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 3 adds Speciale and output tokens; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: browsecomp-cm
deepseek
BrowseComp with the context-management technique (marked *), search agent with a commercial search API; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: browsecompzh
deepseek
browsecompzh; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: gpqa
deepseek
gpqa; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 3 adds Speciale and output tokens; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: hle
deepseek
hle; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 3 adds Speciale and output tokens; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: hle-search
deepseek
hle-search; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: hmmt25
deepseek
hmmt25; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 3 adds Speciale and output tokens; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: hmmtnov
deepseek
hmmtnov; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 3 adds Speciale and output tokens; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: imo
deepseek
imo; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 3 adds Speciale and output tokens; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: lcb
deepseek
lcb; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 3 adds Speciale and output tokens; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: mcpmark
deepseek
mcpmark; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 9 adds non-thinking V3.2; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: mcpuniverse
deepseek
mcpuniverse; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 9 adds non-thinking V3.2; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: mmlupro
deepseek
mmlupro; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: swemulti
deepseek
swemulti; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 9 adds non-thinking V3.2; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: sweverified
deepseek
sweverified; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 9 adds non-thinking V3.2; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: tau2
deepseek
tau2; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 9 adds non-thinking V3.2; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
DeepSeek-V3.2 report: tooldec
deepseek
tooldec; DeepSeek-V3.2 technical report Tables 2/3/9 (arXiv 2512.02556, 2025-12-02; release 2025-12-01); temperature 1.0, context window 128K; math/HLE template; judge/n not stated; Table 9 adds non-thinking V3.2; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
deepseek-v41-techreport: ale
deepseek
Agents' Last Exam (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness official-scaffold(ALE-CLI); all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: autobench
deepseek
AutomationBench v1.0.6 public set (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness official-scaffold; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: babyvision
deepseek
BabyVision w/ tools (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness Claude-Code; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: chartography
deepseek
Chartography w/ tools (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness Claude-Code; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: cybergym
deepseek
CyberGym (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness unstated; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: hle
deepseek
HLE (Pass@1), full set; DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness none(reasoning); all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: hle-tools
deepseek
HLE with tools (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness unstated; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: nl2repo
deepseek
NL2Repo-Bench (Score); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness DeepSeek-Harness-minimal; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: programbench
deepseek
ProgramBench (Almost@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness DeepSeek-Harness-minimal; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: tb30
deepseek
Terminal-Bench 3.0 (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness DeepSeek-Harness-minimal; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: tb40
deepseek
Terminal-Bench 4.0 (Pass@1); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness DeepSeek-Harness-minimal; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
deepseek-v41-techreport: zerobench
deepseek
ZeroBench-main w/ tools (Pass@5); DeepSeek V4.1-Flash tech report / model card; sampling stated by the card: temperature=1.0, top_p=0.95 (report 5.3.1: temperature and top-p 1.0 for reasoning benchmarks); N=8 samples per task for DeepSWE v1.1, N=3 for Terminal-Bench 2.1 (Table 4 note); judge not stated; harness Claude-Code; all columns at max effort for report/card E; 1017 card effort not stated; read 2026-10-19 ✓xmodelKimi K3·max
Design Arena - all categories
designarena
designarena.ai/leaderboard 'allcategories': crowd human-vote Elo on AI-generated designs/UI (1,817,484 votes, 180 models); effort not in the Claude slugs (other vendors' slugs carry -max/-high) so default; no cost; read 2026-10-15 ✓xmodelKimi K2.7 Code·solo
Design Arena - fullstack
designarena
designarena.ai/leaderboard 'fullstack' category: crowd human-vote Elo on full-stack app generation (61,252 votes, 49 models); effort not stated; no cost; read 2026-10-15 ✓xmodelKimi K2.7 Code·solo
Design Creativity Bench (UI generation; overall T-score, median cost per design)
arxiv-2610.11539
80 UI design tasks / 160 prompts, one generation per design, highest effort (Opus 5.5 at xhigh); Tables 2 and 9 of arXiv 2610.11539; read 2026-10-15 ✓xmodelKimi K3·max
dubesor-benchtable
dubesor
—xmodelKimi K2 Thinking·solo
E-Bench (MCP tools): Avg@3 task success
ebench-2607.23722
E-Bench (Tencent-related authors; apps Honor of Kings / QQ Music / Tencent Meeting) E-Bench (MCP tools): 323 tasks, 3 independent trials from fresh DB copies, MCP tool agents with shared harness, deterministic final-DB-diff grader; Avg@3 = mean trial success (Pass@3, Pass^3 same runs); cost = per-task USD computed by the authors from task-average tokens x list price (Table 5; price date not given); arXiv 2607.23722 (28 Jul 2026); read 2026-10-20 ✓xmodelKimi K3·max
E-Bench-Code (code execution): Avg@3 task success
ebench-2607.23722
E-Bench (Tencent-related authors; apps Honor of Kings / QQ Music / Tencent Meeting) E-Bench-Code (code execution): 323 tasks, 3 independent trials from fresh DB copies, MCP tool agents with shared harness, deterministic final-DB-diff grader; Avg@3 = mean trial success (Pass@3, Pass^3 same runs); cost = per-task USD computed by the authors from task-average tokens x list price (Table 5; price date not given); arXiv 2607.23722 (28 Jul 2026); read 2026-10-20 ✓xmodelKimi K3·max
econ-eval leaderboard A (20 tasks, Opus-5 judge)
github-deluair-econ-eval
deluair/econ-eval README at commit 77f9f3e (16 Oct 2026), leaderboard of 2026-09-10: 20 economics tasks (quantitative, reasoning, coding, writing) x 5 samples = 100 completions/model, numeric/exact/code-exec graders + rubric tasks judged by Claude Opus 5; score = mean over tasks of task means; cost = total for the 100 completions from token counts x list price (prices read from the OpenRouter catalog 2026-07-31 / vendor pages); per-track tables are the same completions (not extra groups); ten models ran 2026-08-01, deepseek-flash added 2026-09-10 same tasks/samples/judge; read 2026-10-20 ✓xmodelKimi K3·max
EMB
vals.ai
Vals AI · bash+execute, 10 Claude models at max (payload compute_effort) — snapshot 27 Sep, re-priced ✓xmodelKimi K3·max
Epoch · chess puzzles
epoch.ai
Epoch AI's own evaluation (chess_puzzles), benchmarking hub archive read 30 Sep 2026; effort from the model-version suffix, mean score ✓xmodelKimi K2.7 Code·solo · Kimi K3·max · Kimi K3·high · Kimi K3·low
Epoch · frontiermath tier 4 v2
epoch.ai
Epoch AI's own evaluation (frontiermath_tier_4_v2), benchmarking hub archive read 30 Sep 2026; effort from the model-version suffix, mean score ✓xmodelKimi K3·max · Kimi K2.7 Code·solo
Epoch · frontiermath tiers 1 3 v2
epoch.ai
Epoch AI's own evaluation (frontiermath_tiers_1_3_v2), benchmarking hub archive read 30 Sep 2026; effort from the model-version suffix, mean score ✓xmodelKimi K3·max · Kimi K2.7 Code·solo
Epoch · furniture assembly
epoch.ai
Epoch AI's own evaluation (furniture_assembly), benchmarking hub archive read 30 Sep 2026; effort from the model-version suffix, mean score ✓xmodelKimi K3·max
Epoch · gpqa diamond
epoch.ai
Epoch AI's own evaluation (gpqa_diamond), benchmarking hub archive read 30 Sep 2026; effort from the model-version suffix, mean score ✓xmodelKimi K3·low · Kimi K3·high · Kimi K2.7 Code·solo · Kimi K3·max
Epoch · mystery game puzzles
epoch.ai
Epoch AI's own evaluation (mystery_game_puzzles), benchmarking hub archive read 30 Sep 2026; effort from the model-version suffix, mean score ✓xmodelKimi K3·low · Kimi K2.6·nothink · Kimi K3·max
Epoch · otis mock aime 2024 2025
epoch.ai
Epoch AI's own evaluation (otis_mock_aime_2024_2025), benchmarking hub archive read 30 Sep 2026; effort from the model-version suffix, mean score ✓xmodelKimi K3·low · Kimi K3·high · Kimi K2.7 Code·solo · Kimi K3·max
Epoch · simpleqa verified
epoch.ai
Epoch AI's own evaluation (simpleqa_verified), benchmarking hub archive read 30 Sep 2026; effort from the model-version suffix, mean score ✓xmodelKimi K3·max · Kimi K2.7 Code·solo
epoch-ext-surface-evolver-bench
epoch.ai
—xmodelKimi K3·max · Kimi K2.7 Code·solo
EQ-Bench creative-writing-longform (EQ-bench/eqbench-leaderboard-results)
EQ-bench/eqbench-leaderboard-results (GitHub)
GitHub repo EQ-bench/eqbench-leaderboard-results, pinned commit 266853f2ac7d10c31430cb3bd00d64c758beb581, dir creative-writing-longform/results_sonnet-4.6/, 12/12 multi-chapter longform tasks, judge claude-sonnet-4-6 (flag claude-judge), 1 iteration (n=1); distinct board from the already-held eqbench4 (eqbench.com static roleplay-Elo data, different judge/metric/source) — not a restatement.; read 2026-10-08 ✓xmodelKimi K2 Thinking·solo
fcode20pass
cognition
—xmodelKimi K3·max
fcode20score
cognition
—xmodelKimi K3·max
Financial Touchstone v1.2 accuracy (annual-report QA, shared top-5 retrieval)
arxiv-2608.08634
arXiv 2608.08634 (9 Aug 2026) Financial Touchstone v1.2: 2,967 question-context-answer items on 495 annual reports, same top-5 retrieved context for every model, output cap 5,000 tokens, LLM judge Claude Opus 4.6 (16 models) / Claude Opus 4.8 (Kimi K2.6, GLM-4.7, Qwen3-Max, ERNIE 5.1); score = accuracy excluding retriever errors (Fig. 1); hallucination rate = same runs, other metric (not a group); read 2026-10-20 ✓xmodelKimi K2.6·think
FrontierSWE v2
frontierswe
Proximal frontierswe.com + Sonnet 5.5 card · 34 ultra-long tasks, max effort, proximus harness, mean reward@5; cost basis undocumented ✓xmodelKimi K3·max
gdpx-overall
semianalysis
—xmodelKimi K3·max
gdpx-p-arithmetic
semianalysis
—xmodelKimi K3·max
gdpx-p-beautiful-machines
semianalysis
—xmodelKimi K3·max
gdpx-p-emulated
semianalysis
—xmodelKimi K3·max
gdpx-p-ergodex
semianalysis
—xmodelKimi K3·max
gdpx-p-extrapolate
semianalysis
—xmodelKimi K3·max
gdpx-p-latchbio
semianalysis
—xmodelKimi K3·max
gdpx-p-third-moment
semianalysis
—xmodelKimi K3·max
gdpx-p-valenceforge
semianalysis
—xmodelKimi K3·max
Gert Labs GBENCH agentic coding (models write code run as agents)
gertlabs-gbench
Gert Labs GBENCH global rankings API (gertlabs.com/api/v1/global/rankings?limit=300), mode agentic_coding, gscore 0-1 (board composite of percentile / success / response time), cost = avg_cost_usd per submission as printed by the payload (definition not documented); models' own games set varies by model; live board, payload 2026-10-20T05:55Z; read 2026-10-20 ✓xmodelKimi K2.7 Code·solo
Gert Labs GBENCH decision making (models play the games)
gertlabs-gbench
Gert Labs GBENCH global rankings API (gertlabs.com/api/v1/global/rankings?limit=300), mode decision_making, gscore 0-1 (board composite of percentile / success / response time), cost = avg_cost_usd per submission as printed by the payload (definition not documented); models' own games set varies by model; live board, payload 2026-10-20T05:55Z; read 2026-10-20 ✓xmodelKimi K2.7 Code·solo
Gert Labs GBENCH one-shot coding (models write code in one shot)
gertlabs-gbench
Gert Labs GBENCH global rankings API (gertlabs.com/api/v1/global/rankings?limit=300), mode oneshot_coding, gscore 0-1 (board composite of percentile / success / response time), cost = avg_cost_usd per submission as printed by the payload (definition not documented); models' own games set varies by model; live board, payload 2026-10-20T05:55Z; read 2026-10-20 ✓xmodelKimi K2.7 Code·solo
GPQA-Diamond (GLM-4.7 card)
zhipu
GPQA-Diamond; GLM-4.7 card 2026-01-29; settings: default settings temp 1.0 top-p 0.95 131072; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
hehee9 Maze Bench Tier 1 (30 mazes)
hehee9maze-bcd609c
Maze image (2048x2048) -> single-line move string, per-maze score = 100 x progress x efficiency, 30 mazes (6 sizes x 5, 4x4..18x18), one run per model-maze, single call, image input, no tools, medium effort / 8K budget when supported; metric official mean score 0-100; cost = dashboard calculateCost = (input tokens x input rate + output tokens x output rate)/1e6 from rates stored per model in the JSON (before cache discounts); live board snapshot of 20 Oct 2026 (github.com/hehee9/maze-bench commit bcd609c, status completed); read 2026-10-20 ✓xmodelKimi K3·max
HLE (GLM-4.7 card)
zhipu
HLE; GLM-4.7 card 2026-01-29; settings: default settings; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
HLE w/ tools (GLM-4.7 card)
zhipu
HLE w/ tools; GLM-4.7 card 2026-01-29; settings: default settings; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
HMMT Feb 2025 (GLM-4.7 card)
zhipu
HMMT Feb 2025; GLM-4.7 card 2026-01-29; settings: default settings; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
Hy-MultiTurn (Chinese deep multi-turn dialogue understanding): Importance score
hymultiturn-2607.29196
Hy-MultiTurn leaderboard (Tencent Hunyuan authors; Hy3/Doubao rows are not in the eight providers): six dimensions, final-turn response scored against weighted rubric items; Importance (weighted correct share) = primary metric, mean of three independent rolls; arXiv 2607.29196 (31 Jul 2026); read 2026-10-20 ✓xmodelKimi K3·max · Kimi K2.6·think · Kimi K2.6·nothink
hypfrontier-challenge64-hf-valid
hypfrontier-2608.10843
—xmodelKimi K2.7 Code·solo
ibmartian-full-p-other
ironbee
—xmodelKimi K3·low
InferenceBench (prompt v1.0.1 (†))
inferencebench.ai
InferenceBench, agents tuning LLM inference serving in Claude Code · geometric-mean speed-up over the PyTorch eager baseline across four scenarios, ±SEM over seed-pair runs in the flags; runs with the benchmark prompt v1.0.1 (†) only; effort from the model name; no cost; read 2026-09-30 ✓xmodelKimi K2.7 Code·solo
ITBench-AA (SRE)
artificialanalysis
Artificial Analysis ITBench-AA (SRE); score only from the model-page payload; effort from payload effort.slug; read 2026-10-15 ✓xmodelKimi K2.6·think · Kimi K3·max
K3 report Fig.13b: BrowseComp score vs cost per task
moonshot
BrowseComp; K3 with context compaction at 300K tokens; GPT-5.6 Sol per effort medium/high/max (points read from the vector drawing of Fig.13b); cost USD per task as plotted (source of costs not stated); K3 tech report p.32; read 2026-10-19 ✓xmodelKimi K3·max
K3 report Table 3: Agent Behavior Bench
moonshot
Agent Behavior Bench (Moonshot in-house); harness Kimi Work; max reasoning effort (GPT-5.5 xhigh); K3 tech report Table 3 (p.29), not on the model card; read 2026-10-19 ✓xmodelKimi K3·max
K3 report Table 3: Chat All-in-One Bench
moonshot
Chat All-in-One Bench (Moonshot in-house); harness Kimi Work; max reasoning effort (GPT-5.5 xhigh); K3 tech report Table 3 (p.29), not on the model card; read 2026-10-19 ✓xmodelKimi K3·max
K3 report Table 3: DECK Bench
moonshot
DECK Bench (Moonshot in-house); harness N/A; max reasoning effort (GPT-5.5 xhigh); K3 tech report Table 3 (p.29), not on the model card; read 2026-10-19 ✓xmodelKimi K3·max
K3 report Table 3: Faithfulness
moonshot
Faithfulness (Moonshot in-house); harness N/A; max reasoning effort (GPT-5.5 xhigh); K3 tech report Table 3 (p.29), not on the model card; read 2026-10-19 ✓xmodelKimi K3·max
K3 report Table 3: Finance Bench
moonshot
Finance Bench (Moonshot in-house); harness N/A; max reasoning effort (GPT-5.5 xhigh); K3 tech report Table 3 (p.29), not on the model card; read 2026-10-19 ✓xmodelKimi K3·max
K3 report Table 3: KWV Bench
moonshot
KWV Bench (Moonshot in-house); harness N/A; max reasoning effort (GPT-5.5 xhigh); K3 tech report Table 3 (p.29), not on the model card; read 2026-10-19 ✓xmodelKimi K3·max
K3 report Table 3: MIRA Bench
moonshot
MIRA Bench (Moonshot in-house); harness MIRA; max reasoning effort (GPT-5.5 xhigh); K3 tech report Table 3 (p.29), not on the model card; read 2026-10-19 ✓xmodelKimi K3·max
KernelBench MEGA (Claude Code)
github-kernelbench
02_kimi_linear_decode megakernel on RTX PRO 6000 Blackwell · harness=claude (Claude Code) · n=1 per cell, runs months apart (Jun-Sep 2026); NOT the board's CUDA task (that is the held kernelbench-cc/or group); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
Kimi K2.6 card: AIME 2026
moonshot
AIME 2026; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: APEX-Agents
moonshot
APEX-Agents; 452 of the 480 public tasks (as Artificial Analysis); temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: BabyVision
moonshot
BabyVision; vision; max tokens 98,304; avg@3; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: BabyVision (w/ python)
moonshot
BabyVision (w/ python); vision; max tokens 98,304; avg@3; Python tool: 65,536 per step, 50 steps; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: BrowseComp (Agent Swarm)
moonshot
BrowseComp (Agent Swarm); discard-all context management; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: CharXiv (RQ)
moonshot
CharXiv (RQ); vision; max tokens 98,304; avg@3; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: CharXiv (RQ) (w/ python)
moonshot
CharXiv (RQ) (w/ python); vision; max tokens 98,304; avg@3; Python tool: 65,536 per step, 50 steps; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: Claw Eval (pass^3)
moonshot
Claw Eval (pass^3); Claw Eval v1.1; max tokens per step 16,384; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: DeepSearchQA (f1-score)
moonshot
DeepSearchQA (f1-score); temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: HLE-Full
moonshot
HLE-Full; full set; max generation 98,304 tokens; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: HLE-Full (w/ tools)
moonshot
HLE-Full (w/ tools); full set; max generation 98,304 tokens; tools: search/code/browse; per-step 49,152; context management; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: HMMT 2026 (Feb)
moonshot
HMMT 2026 (Feb); temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: IMO-AnswerBench
moonshot
IMO-AnswerBench; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: LiveCodeBench (v6)
moonshot
LiveCodeBench (v6); temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: MathVision
moonshot
MathVision; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: MathVision (w/ python)
moonshot
MathVision (w/ python); temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: MCPMark
moonshot
MCPMark; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: MMMU-Pro
moonshot
MMMU-Pro; vision; max tokens 98,304; avg@3; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: MMMU-Pro (w/ python)
moonshot
MMMU-Pro (w/ python); vision; max tokens 98,304; avg@3; Python tool: 65,536 per step, 50 steps; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: OJBench (python)
moonshot
OJBench (python); temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: OSWorld-Verified
moonshot
OSWorld-Verified; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: SciCode
moonshot
SciCode; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: SWE-Bench Multilingual
moonshot
SWE-Bench Multilingual; in-house SWE-agent-derived framework; avg of 10 runs; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: Toolathlon
moonshot
Toolathlon; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.6 card: V* (w/ python)
moonshot
V* (w/ python); vision; max tokens 98,304; avg@3; Python tool: 65,536 per step, 50 steps; temperature 1.0, top-p 1.0, context 262,144; Kimi K2.6 model card (HF created 2026-04-14); read 2026-10-19 ✓xmodelKimi K2.6·think
Kimi K2.7 Code card: Kimi Claw 24/7 Bench
moonshot
Kimi Claw 24/7 Bench; in-house; 17 scenarios/610 evaluation points; OpenClaw harness; avg of 3 runs; general setting: temperature 1.0, top-p 0.95, 262,144 context; Kimi K2.7 Code model card (HF created 2026-06-11); read 2026-10-19 ✓xmodelKimi K2.7 Code·solo
Kimi K2.7 Code card: Kimi Code Bench v2
moonshot
Kimi Code Bench v2; in-house; 10+ languages; general setting: temperature 1.0, top-p 0.95, 262,144 context; Kimi K2.7 Code model card (HF created 2026-06-11); read 2026-10-19 ✓xmodelKimi K2.7 Code·solo
Kimi K2.7 Code card: MCP Atlas
moonshot
MCP Atlas; 100 tool-call budget; 32k max tokens per step; avg of 3 runs; general setting: temperature 1.0, top-p 0.95, 262,144 context; Kimi K2.7 Code model card (HF created 2026-06-11); read 2026-10-19 ✓xmodelKimi K2.7 Code·solo
Kimi K2.7 Code card: MCP Mark Verified
moonshot
MCP Mark Verified; human-verified MCPMark; 100-step tool-call budget; 32k max tokens per step; avg of 3 runs; general setting: temperature 1.0, top-p 0.95, 262,144 context; Kimi K2.7 Code model card (HF created 2026-06-11); read 2026-10-19 ✓xmodelKimi K2.7 Code·solo
Kimi K2.7 Code card: MLS Bench Lite
moonshot
MLS Bench Lite; 30-task subset; 5 h per task; general setting: temperature 1.0, top-p 0.95, 262,144 context; Kimi K2.7 Code model card (HF created 2026-06-11); read 2026-10-19 ✓xmodelKimi K2.7 Code·solo
Kimi K2.7 Code card: Program Bench
moonshot
Program Bench; 200 tasks; 248k fuzz tests; general setting: temperature 1.0, top-p 0.95, 262,144 context; Kimi K2.7 Code model card (HF created 2026-06-11); read 2026-10-19 ✓xmodelKimi K2.7 Code·solo
Kimi K3 card: AA-Briefcase (Elo)
moonshot
AA-Briefcase Elo; board snapshot as of 2026-07-23; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: AA-LCR
moonshot
AA-LCR; board snapshot as of 2026-07-23; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: APEX-Agents
moonshot
APEX-Agents; Mercor leaderboard as of 2026-07-23; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: AutomationBench
moonshot
AutomationBench; 600-task public subset; official GitHub setup; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: BabyVision w/ python
moonshot
BabyVision with Python; avg of 3 runs; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: CharXiv (RQ) (no tools)
moonshot
CharXiv reasoning questions; avg of 3 runs; no tools; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: CharXiv (RQ) (with python)
moonshot
CharXiv reasoning questions; avg of 3 runs; with python; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: GDPval-AA v2 (Elo)
moonshot
GDPval-AA v2 Elo; board snapshot as of 2026-07-23; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: GPQA Diamond
moonshot
GPQA Diamond; no tools; top-p 0.95; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 0.95 for single-step tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: HLE-Full (no tools)
moonshot
HLE full set; top-p 0.95; with-tools = general tools; no tools; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 0.95 for single-step tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: HLE-Full (with tools)
moonshot
HLE full set; top-p 0.95; with-tools = general tools; with tools; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: MathVision (no tools)
moonshot
MathVision; avg of 3 runs; no tools; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 0.95 for single-step tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: MathVision (with python)
moonshot
MathVision; avg of 3 runs; with python; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 0.95 for single-step tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: MCP-Atlas
moonshot
MCP-Atlas; 500-task public subset; 100-turn limit; judge Gemini 3.1 Pro; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: MCPMark-Verified
moonshot
MCPMark-Verified; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: MMMU-Pro (no tools)
moonshot
MMMU-Pro official protocol (input order kept; images first); avg of 3 runs; no tools; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: MMMU-Pro (with python)
moonshot
MMMU-Pro official protocol (input order kept; images first); avg of 3 runs; with python; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: MMVU
moonshot
MMVU; avg of 3 runs; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: OmniDocBench
moonshot
OmniDocBench; avg of 3 runs; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: OSWorld 2.0
moonshot
OSWorld 2.0; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: OSWorld-Verified
moonshot
OSWorld-Verified; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: PerceptionBench
moonshot
PerceptionBench (Moonshot in-house benchmark); avg of 3 runs; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: ResearchRubrics
moonshot
ResearchRubrics; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: SaaS-Bench
moonshot
SaaS-Bench; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: SciCode
moonshot
SciCode; board snapshot as of 2026-07-23; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: Toolathlon-Verified
moonshot
Toolathlon-Verified; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: Video-MME (w. sub)
moonshot
Video-MME with subtitles; avg of 3 runs; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: WorldVQA ForceAnswer
moonshot
WorldVQA ForceAnswer variant; vision avg of 3 runs; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: ZeroBench (pass@5) (no tools)
moonshot
ZeroBench main set; pass@5 (run 5 times, official setting); no tools; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: ZeroBench (pass@5) (with python)
moonshot
ZeroBench main set; pass@5 (run 5 times, official setting); with python; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
Kimi K3 card: τ³-Banking
moonshot
tau3-Banking; board snapshot as of 2026-07-23; Moonshot Kimi K3 model card table 2026-07 (card snapshot read; HF lastModified 2026-09-02); reasoning effort max, temperature 1.0, top-p 1.0 for agentic tasks; read 2026-10-19 ✓xmodelKimi K3·max
latitude-sweep
github-latitude
—xmodelKimi K3·max
lb2-browser-analysis-driven-action
livebench
—xmodelKimi K3·max
lb2-browser-constraint-adherence
livebench
—xmodelKimi K3·max
lb2-browser-data-aggregation
livebench
—xmodelKimi K3·max
lb2-browser-sequential-planning
livebench
—xmodelKimi K3·max
lb2-browser-structural-integrity
livebench
—xmodelKimi K3·max
lb2-coding-billing-limits
livebench
—xmodelKimi K3·max
lb2-coding-cloud-infrastructure
livebench
—xmodelKimi K3·max
lb2-coding-hard-fixes
livebench
—xmodelKimi K3·max
lb2-coding-llm-api-integration
livebench
—xmodelKimi K3·max
lb2-coding-memory-tools
livebench
—xmodelKimi K3·max
lb2-coding-product-ux
livebench
—xmodelKimi K3·max
lb2-dataanalysis-consecutive-events
livebench
—xmodelKimi K3·max
lb2-dataanalysis-table-join
livebench
—xmodelKimi K3·max
lb2-dataanalysis-table-reformat
livebench
—xmodelKimi K3·max
lb2-design-documents-docx
livebench
—xmodelKimi K3·max
lb2-design-pdf
livebench
—xmodelKimi K3·max
lb2-design-presentations-ppt
livebench
—xmodelKimi K3·max
lb2-longhorizon-delegated-operations
livebench
—xmodelKimi K3·max
lb2-longhorizon-finance-compliance-apps
livebench
—xmodelKimi K3·max
lb2-longhorizon-long-run-operations
livebench
—xmodelKimi K3·max
lb2-longhorizon-operations-commerce-apps
livebench
—xmodelKimi K3·max
lb2-longhorizon-performance-optimisation
livebench
—xmodelKimi K3·max
lb2-longhorizon-systems-engineering
livebench
—xmodelKimi K3·max
lb2-reasoning-logic-with-navigation
livebench
—xmodelKimi K3·max
lb2-reasoning-theory-of-mind
livebench
—xmodelKimi K3·max
lb2-research-legal
livebench
—xmodelKimi K3·max
lb2-research-regulatory-compliance
livebench
—xmodelKimi K3·max
lb2-research-science-evidence
livebench
—xmodelKimi K3·max
lb2-research-strategy
livebench
—xmodelKimi K3·max
lb2-research-tax-finance
livebench
—xmodelKimi K3·max
Lechmazur Elimination Game (TrueSkill mu)
lechmazur-elimination-game
multi-player elimination tournament (public/private chat, votes, jury), TrueSkill mu aggregated over random passes; games per model printed; README leaderboard of 61 rows; changelog 'Jan 6, 2026: GPT-5.2, Opus 4.5, Gemini 3 Pro, Gemini 3 Flash, Kimi K2 Thinking, Qwen 3 Max Thinking, Grok 4.1 Fast ... added' (stale board); no cost; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
LiveBench 2.0 · Browser Use suite
livebench
LiveBench 2.0 Browser Use suite, single run, every model at its highest effort (max), cost = suite total at list price; read 2026-10-15 ✓xmodelKimi K3·max
LiveBench 2.0 · Coding suite
livebench
LiveBench 2.0 Coding suite, single run, every model at its highest effort (max), cost = suite total at list price; read 2026-10-15 ✓xmodelKimi K3·max
LiveBench 2.0 · Data Analysis suite
livebench
LiveBench 2.0 Data Analysis suite, single run, every model at its highest effort (max), cost = suite total at list price; read 2026-10-15 ✓xmodelKimi K3·max
LiveBench 2.0 · Design suite
livebench
LiveBench 2.0 Design suite, single run, every model at its highest effort (max), cost = suite total at list price; read 2026-10-15 ✓xmodelKimi K3·max
LiveBench 2.0 · Long-Horizon suite
livebench
LiveBench 2.0 Long-Horizon suite, single run, every model at its highest effort (max), cost = suite total at list price; read 2026-10-15 ✓xmodelKimi K3·max
LiveBench 2.0 · Reasoning suite
livebench
LiveBench 2.0 Reasoning suite, single run, every model at its highest effort (max), cost = suite total at list price; read 2026-10-15 ✓xmodelKimi K3·max
LiveBench 2.0 · Research suite
livebench
LiveBench 2.0 Research suite, single run, every model at its highest effort (max), cost = suite total at list price; read 2026-10-15 ✓xmodelKimi K3·max
livebench-task-amps_hard
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-code_completion
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-code_generation
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-connections
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-consecutive_events
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-integrals_with_game
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-javascript
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-logic_with_navigation
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-math_comp
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-olympiad
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-paraphrase
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-plot_unscrambling
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-python
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-simplify
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-spatial
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-story_generation
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-summarize
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-tablejoin
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-tablereformat
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-theory_of_mind
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-typescript
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-typos
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
livebench-task-zebra_puzzle
livebench
—xmodelKimi K2 Thinking·solo · Kimi K2.6·think
LiveCodeBench-v6 (GLM-4.7 card)
zhipu
LiveCodeBench-v6; GLM-4.7 card 2026-01-29; settings: default settings; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
llm-stats.com arena: agents category (Bradley-Terry/TrueSkill-style, own-run)
llmstats.com
llm-stats.com's own per-category arena rating (mu/sigma/CI/games_played), agents category, methodology not fully disclosed on homepage; read 2026-10-05 ✓xmodelKimi K2.7 Code·solo
llm-stats.com arena: general category (Bradley-Terry/TrueSkill-style, own-run)
llmstats.com
llm-stats.com's own per-category arena rating (mu/sigma/CI/games_played), general category, methodology not fully disclosed on homepage; read 2026-10-05 ✓xmodelKimi K2.7 Code·solo
llm-stats.com arena: reasoning category (Bradley-Terry/TrueSkill-style, own-run)
llmstats.com
llm-stats.com's own per-category arena rating (mu/sigma/CI/games_played), reasoning category, methodology not fully disclosed on homepage; read 2026-10-05 ✓xmodelKimi K2.7 Code·solo
llm2014-logic-2026-10-median
llm2014-logic
—xmodelKimi K3·max
LLMConfBench
arxiv-2609.20666
arXiv 2609.20666 · conformer energy ranking, 27 molecules × 3 prompts via OpenRouter, Opus 5 and Sonnet 5 at high; cost summed from the repo's per-call logs ✓xmodelKimi K3·high
LMArena Agent
lmarena.ai
Agent Mode real sessions (Code/Chat/Work) · cost=median $/task (14d window), score=Net Improvement composite signal, effort read from model-variant label; live board, read 27 Sep 2026 ✓xmodelKimi K3·max
LMArena Text
lmarena.ai
Text arena (Overall) · blind pairwise votes, Elo rating, no cost; effort read from the arena's model key (e.g. claude-opus-5.5-high-vertex); live board, read 30 Sep 2026 ✓xmodelKimi K3·max · Kimi K2 Thinking·solo
LMArena Text (xAI Grok 4.1 post)
xai
LMArena Text Arena Elo as printed by xAI (chart label 'Overall Style Control Elo', text 'overall'; style control unclear), live board snapshot of 17 Nov 2025; no cost; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
LMArena WebDev
lmarena.ai
Code arena, WebDev board (Overall) · blind pairwise votes, Elo rating, no cost; effort read from the arena's model key (e.g. claude-sonnet-5-5-high); live board, read 30 Sep 2026 ✓xmodelKimi K3·max · Kimi K2.7 Code·solo · Kimi K2 Thinking·solo
LMArena WebDev — HTML sub-category
lmarena.ai
lmarena-ai/leaderboard-dataset, configs/webdev, category=webdev-html, latest split, publish 2026-10-02; read 2026-10-05 ✓xmodelKimi K3·max · Kimi K2.7 Code·solo · Kimi K2 Thinking·solo
LMArena WebDev — React sub-category
lmarena.ai
lmarena-ai/leaderboard-dataset, configs/webdev, category=webdev-react, latest split, publish 2026-10-02; read 2026-10-05 ✓xmodelKimi K3·max · Kimi K2.7 Code·solo · Kimi K2 Thinking·solo
MathArena · aime 2026
matharena.ai
MathArena, aime--aime_2026: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max · Kimi K2.6·think
MathArena · apex 2025
matharena.ai
MathArena, apex--apex_2025: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max · Kimi K2.6·think
MathArena · april
matharena.ai
MathArena, arxiv--april: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · april
matharena.ai
MathArena, arxiv_false--april: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · august
matharena.ai
MathArena, arxiv--august: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · august
matharena.ai
MathArena, arxiv_false--august: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · february
matharena.ai
MathArena, arxiv--february: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K2.6·think
MathArena · february
matharena.ai
MathArena, arxiv_false--february: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K2.6·think
MathArena · hmmt feb 2026
matharena.ai
MathArena, hmmt--hmmt_feb_2026: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max · Kimi K2.6·think
MathArena · january
matharena.ai
MathArena, arxiv--january: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K2.6·think
MathArena · june
matharena.ai
MathArena, arxiv--june: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · june
matharena.ai
MathArena, arxiv_false--june: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · kangaroo 2025 1-2
matharena.ai
MathArena, kangaroo--kangaroo_2025_1-2: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · kangaroo 2025 11-12
matharena.ai
MathArena, kangaroo--kangaroo_2025_11-12: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · kangaroo 2025 3-4
matharena.ai
MathArena, kangaroo--kangaroo_2025_3-4: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · kangaroo 2025 5-6
matharena.ai
MathArena, kangaroo--kangaroo_2025_5-6: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · kangaroo 2025 7-8
matharena.ai
MathArena, kangaroo--kangaroo_2025_7-8: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · march
matharena.ai
MathArena, arxiv--march: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K2.6·think
MathArena · march
matharena.ai
MathArena, arxiv_false--march: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K2.6·think
MathArena · may
matharena.ai
MathArena, arxiv--may: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · may
matharena.ai
MathArena, arxiv_false--may: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max
MathArena · shortlist 2025
matharena.ai
MathArena, apex--shortlist_2025: each model run 4 times on each problem · accuracy in % with its 95 % CI in the flags, cost = the average cost of one run on one problem (USD), mean input/output tokens; effort from the model name; ⚠️ (model released after the competition) kept as a flag; read 2026-09-30 ✓xmodelKimi K3·max · Kimi K2.6·think
Mercor · Harvey LAB
mercor.com
Mercor, Harvey LAB: the open-source benchmark independently evaluated by Mercor · score in %, ± kept in the flags; the card shows its top 5 models only; effort from the board's per-model field; no cost; read 2026-09-30 ✓xmodelKimi K3·max
Mercor · Long Context Reasoning (AA-LCR v1.1)
mercor.com
Mercor, Long Context Reasoning (AA-LCR v1.1): the open-source benchmark independently evaluated by Mercor · score in %, ± kept in the flags; the card shows its top 5 models only; effort from the board's per-model field; no cost; read 2026-09-30 ✓xmodelKimi K3·max
Mercor · SWE Atlas - Codebase QnA
mercor.com
Mercor, SWE Atlas - Codebase QnA: the open-source benchmark independently evaluated by Mercor · score in %, ± kept in the flags; the card shows its top 5 models only; effort from the board's per-model field; no cost; read 2026-09-30 ✓xmodelKimi K3·max
Mercor APEX-Accounting
mercor.com
Mercor's own board, APEX-Accounting, harness loop_agent · pass@1 in %, Mercor's ± half-width kept in the flags, mean score beside it; effort from the board's per-model field; no cost; read 2026-09-30 ✓xmodelKimi K3·max · Kimi K2.7 Code·solo
Mercor APEX-Agents
mercor.com
Mercor's own board, APEX-Agents, harness loop_truncated_tools_agent · pass@1 in %, Mercor's ± half-width kept in the flags, mean score beside it; effort from the board's per-model field; no cost; read 2026-09-30 ✓xmodelKimi K3·max · Kimi K2.7 Code·solo
Mercor APEX-v1 (The AI Productivity Index)
mercor.com
Mercor's own board, APEX-v1 (The AI Productivity Index), harness singleshot_agent · pass@1 in %, Mercor's ± half-width kept in the flags, mean score beside it; effort from the board's per-model field; no cost; read 2026-09-30 ✓xmodelKimi K2.7 Code·solo · Kimi K3·max
mercor-apexacct-mean
mercor.com
—xmodelKimi K3·max · Kimi K2.7 Code·solo
mercor-apexv1-mean
mercor.com
—xmodelKimi K2.7 Code·solo · Kimi K3·max
MMLU-Pro (GLM-4.7 card)
zhipu
MMLU-Pro; GLM-4.7 card 2026-01-29; settings: default settings temp 1.0 top-p 0.95 131072; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
Next.js Agent Evals (Vercel), base run
nextjs-evals
31 Next.js agent evals, 4 runs each (earlyExit), Claude Code via AI Gateway, no AGENTS.md; https://nextjs.org/evals, github.com/vercel/next-evals-oss@38cb8ad, agent-results.json metadata.experiments[].baseSuccessRate/avgCostUsd; read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
Next.js Agent Evals (Vercel), with AGENTS.md
nextjs-evals
Same 31 evals/4 runs/Claude Code setup as nextjsevals-base, but with an AGENTS.md/CLAUDE.md file telling the agent the Next.js version has breaking changes; score = docsSuccessRate (board); cost = mean of per-eval costUsd in agent-results.json['results'][model] (computed-by-us, since the board's avgCostUsd field is the base-run figure, not the AGENTS.md-run figure — the per-eval array's own mean, 0.2456 for Opus 5.5, differs from the base 0.234, confirming the two are different runs); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
NonoBench
nonobench.com
NonoBench, 30 nonogram puzzles · puzzles solved per effort level (the board's effort-ladder chart); total cost of the 30 puzzles for each model's best configuration (leaderboard rows); Sonnet 4.5 reasoning off/on as nothink/default; read 2026-09-30 ✓xmodelKimi K3·max
Nuxt agent evals (nuxt/nuxt-evals), Claude Code via AI Gateway
nuxt/nuxt-evals
agent-results.json exportedAt 2026-09-22; 31 Nuxt agent evals, runs: 4, earlyExit: true; score is avgPassRate (fraction of the 4 runs passing, averaged over 31 evals), *100 for a percentage; cost is the gateway's avg cost USD per eval run; models ran on different dates (opus-5.5 2026-09-22, the rest 2026-08-27 to 09-05); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
nuxt-evals-success
nuxt/nuxt-evals
—xmodelKimi K2.7 Code·solo
NYT Connections, one submission per puzzle, 20 puzzles (matsonj/eval-connections)
matsonj/eval-connections
results.md (data behind matsonj.github.io/eval-connections); pts out of max 100 (20 puzzles x up to 5 pts, wins/traps/groups also recorded); cost is total USD for the 20-game run; models ran on different dates (see confound); single run each; read 2026-10-01 ✓xmodelKimi K3·low
OpenRouter GPQA-Diamond rolling-mean board (snapshot 2026-10-03)
openrouter
reasoning-math/hard, OR-internal rolling-mean harness, auto-routing provider aggregate; ref openrouter.ai/benchmarks/gpqa-diamond; SNAPSHOT REPLACES prior orgpqa snapshot (2026-09-27); read 2026-10-05 ✓xmodelKimi K2 Thinking·solo · Kimi K2.7 Code·solo
OpenRouter tau2-bench-airline board (snapshot 2026-10-03)
openrouter
agentic-tool/mid, tau2-bench airline tasks, OR-internal harness, auto-routing provider aggregate; ref openrouter.ai/benchmarks/tau2-bench-airline; SNAPSHOT REPLACES prior ortau2 snapshot (2026-09-27); read 2026-10-05 ✓xmodelKimi K2.7 Code·solo · Kimi K2 Thinking·solo
openvl-base-avg22
arxiv-2610.26860
—xmodelKimi K3·max
openvl-ours-avg22
arxiv-2610.26860
—xmodelKimi K3·max
OSWorld-Pro
arxiv-2609.24890
arXiv 2609.24890 · OSWorld harness, 250 turns, 4 Claude models all at max, tokens × OpenRouter list incl. cache ✓xmodelKimi K3·max
prinzbench (private legal-research + search benchmark by a single author), leaderboard of 18 Jul 2026
github-prinz-ai-prinzbench
prinzbench (github.com/prinz-ai/prinzbench README 'Last updated on July 18, 2026'): 33 author-written questions (25 legal research + 8 needle search) asked 3 times, pass@1, graded by the author without blinding; score = full score x/99; access through chat apps or terminals at different dates (Jan-Jul 2026); '*' = early access before public release; the author cautions that Pro models use parallel test-time compute; read 2026-10-20 ✓xmodelKimi K3·max
reactbench
reactbench
—xmodelKimi K2.7 Code·solo
reactbench-fix
reactbench
—xmodelKimi K2.7 Code·solo
reactbench-write
reactbench
—xmodelKimi K2.7 Code·solo
ReLE 专业能力 (professional ability)
chinese-llm-benchmark
jeinlee1991/chinese-llm-benchmark leaderboard/专业能力.md (专业能力); 171-leaderboard Chinese-language board; score=accuracy%, cost=CNY per 1,000 calls (花费/千次(元)); no effort level stated by the source (plain '-thinking' suffix vs plain model name is the only variant given); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
ReLE 医疗与心理健康 (medical/mental health)
chinese-llm-benchmark
jeinlee1991/chinese-llm-benchmark leaderboard/医疗与心理健康.md (医疗与心理健康); 171-leaderboard Chinese-language board; score=accuracy%, cost=CNY per 1,000 calls (花费/千次(元)); no effort level stated by the source (plain '-thinking' suffix vs plain model name is the only variant given); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
ReLE 总分 (total)
chinese-llm-benchmark
jeinlee1991/chinese-llm-benchmark leaderboard/total.md (alldata); 171-leaderboard Chinese-language board; score=accuracy%, cost=CNY per 1,000 calls (花费/千次(元)); no effort level stated by the source (plain '-thinking' suffix vs plain model name is the only variant given); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
ReLE 推理与数学计算 (reasoning/math)
chinese-llm-benchmark
jeinlee1991/chinese-llm-benchmark leaderboard/推理与数学计算.md (推理与数学计算); 171-leaderboard Chinese-language board; score=accuracy%, cost=CNY per 1,000 calls (花费/千次(元)); no effort level stated by the source (plain '-thinking' suffix vs plain model name is the only variant given); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
ReLE 教育 (education)
chinese-llm-benchmark
jeinlee1991/chinese-llm-benchmark leaderboard/教育.md (教育); 171-leaderboard Chinese-language board; score=accuracy%, cost=CNY per 1,000 calls (花费/千次(元)); no effort level stated by the source (plain '-thinking' suffix vs plain model name is the only variant given); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
ReLE 法律与行政公务 (law/admin)
chinese-llm-benchmark
jeinlee1991/chinese-llm-benchmark leaderboard/法律与行政公务.md (法律与行政公务); 171-leaderboard Chinese-language board; score=accuracy%, cost=CNY per 1,000 calls (花费/千次(元)); no effort level stated by the source (plain '-thinking' suffix vs plain model name is the only variant given); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
ReLE 语言与指令遵从 (language/IF)
chinese-llm-benchmark
jeinlee1991/chinese-llm-benchmark leaderboard/语言与指令遵从.md (语言与指令遵从); 171-leaderboard Chinese-language board; score=accuracy%, cost=CNY per 1,000 calls (花费/千次(元)); no effort level stated by the source (plain '-thinking' suffix vs plain model name is the only variant given); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
ReLE 通用能力 (general ability)
chinese-llm-benchmark
jeinlee1991/chinese-llm-benchmark leaderboard/通用能力.md (通用能力); 171-leaderboard Chinese-language board; score=accuracy%, cost=CNY per 1,000 calls (花费/千次(元)); no effort level stated by the source (plain '-thinking' suffix vs plain model name is the only variant given); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
ReLE 金融 (finance)
chinese-llm-benchmark
jeinlee1991/chinese-llm-benchmark leaderboard/金融.md (金融); 171-leaderboard Chinese-language board; score=accuracy%, cost=CNY per 1,000 calls (花费/千次(元)); no effort level stated by the source (plain '-thinking' suffix vs plain model name is the only variant given); read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
ReplaySCM Block-order (100 problems): HeldoutWorldExact
replayscm-2605.08197
ReplaySCM Table 1 Block-order (100 problems), executable Boolean SCM induction, direct generation no tools, one stored response per model/problem, temp 0.1 (0.1 where set); metric HeldoutWorldExact (other columns = same runs, rule 8); arXiv 2605.08197 (5 May 2026); read 2026-10-20 ✓xmodelKimi K2 Thinking·solo
ReplaySCM Hidden-order Hid-Full (250 problems): HeldoutWorldExact
replayscm-2605.08197
ReplaySCM Table 1 Hidden-order Hid-Full (250 problems), executable Boolean SCM induction, direct generation no tools, one stored response per model/problem, temp 0.1 (0.1 where set); metric HeldoutWorldExact (other columns = same runs, rule 8); arXiv 2605.08197 (5 May 2026); read 2026-10-20 ✓xmodelKimi K2 Thinking·solo
ReplaySCM Ord-Full (250 problems): HeldoutWorldExact
replayscm-2605.08197
ReplaySCM Table 1 Ord-Full (250 problems), executable Boolean SCM induction, direct generation no tools, one stored response per model/problem, temp 0.1 (0.1 where set); metric HeldoutWorldExact (other columns = same runs, rule 8); arXiv 2605.08197 (5 May 2026); read 2026-10-20 ✓xmodelKimi K2 Thinking·solo
Roboflow Playground Vision Evals, OCR (low tier)
roboflow-playground-evals
Roboflow Playground Vision Evals (playground.roboflow.com/evals), 62-63 VLMs, 5 real-world tasks scored against ground truth by Roboflow, scores pooled at the low effort tier (page: 'scores and efficiency pooled across all five tasks at low effort'); page updated 2026-10-13, pricing 2026-10-19; read 2026-10-20 via r.jina.ai; task column: OCR; only cells that differ from the held vlm-exam cells (github-vlmexam, group vlmexam-ocr) or models the held board never ran are kept; no cost per task; read 2026-10-20 ✓xmodelKimi K2.6·nothink
Roboflow Playground Vision Evals, overall (mean of tasks) with cost per sample
roboflow-playground-evals
Roboflow Playground Vision Evals (playground.roboflow.com/evals), 62-63 VLMs, 5 real-world tasks scored against ground truth by Roboflow, scores pooled at the low effort tier (page: 'scores and efficiency pooled across all five tasks at low effort'); page updated 2026-10-13, pricing 2026-10-19; read 2026-10-20 via r.jina.ai; Overall = page's mean over the tasks scored (4/5 where OCR is missing); cost = 'Avg cost / sample' in USD from price x tokens at the pricing date 2026-10-19 ('~' = estimated from self-host/OpenRouter pricing); read 2026-10-20 ✓xmodelKimi K2.6·nothink
scorch-bench: artillery duel game vs 7 bots, Claude Code subagents (reasoning track), seed 17
github-scorch-bench
smerwin/scorch-bench bench/RESULTS-claude.md (2026-10-09); 8 matches per model on one seed; one fresh Claude Code subagent per round via relay provider; score = win rate %; read 2026-10-15 ✓xmodelKimi K3·max
SenseBench (WSD) · prompt p001
sense-bench.com
SenseBench lexEN v1, word-sense disambiguation, prompt set p001 · accuracy in % (± in the flags), cost in USD per million items as the board prices it; effort from the model name; read 2026-09-30 ✓xmodelKimi K2.7 Code·solo
SimpleBench multiple-choice (AVG@5) with cost per run
simplebench
SimpleBench >200 private multiple-choice questions (spatio-temporal, social intelligence, trick questions); score = Score (AVG@5) as printed; cost = modelMeta.cpr 'pre-computed cost to run the whole benchmark once (measured tokens x OpenRouter price)' USD; tout = measured output tokens (unit unstated); live board, rows added between 2025-10 and 2026-10-01 (dateAdded printed); read 2026-10-19 ✓xmodelKimi K3·max · Kimi K2.7 Code·solo · Kimi K2 Thinking·solo
SuperCLUE 2025年度测评 代码生成
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2025年度测评, column 代码生成, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
SuperCLUE 2025年度测评 幻觉控制
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2025年度测评, column 幻觉控制, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
SuperCLUE 2025年度测评 数学推理
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2025年度测评, column 数学推理, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
SuperCLUE 2025年度测评 智能体(任务规划)
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2025年度测评, column 智能体(任务规划), closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
SuperCLUE 2025年度测评 科学推理
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2025年度测评, column 科学推理, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
SuperCLUE 2025年度测评 精确指令遵循
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2025年度测评, column 精确指令遵循, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
SuperCLUE 2026年5月 代码生成
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年5月, column 代码生成, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2.6·think
SuperCLUE 2026年5月 幻觉控制
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年5月, column 幻觉控制, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2.6·think
SuperCLUE 2026年5月 数学推理
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年5月, column 数学推理, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2.6·think
SuperCLUE 2026年5月 智能体(任务规划)
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年5月, column 智能体(任务规划), closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2.6·think
SuperCLUE 2026年5月 科学推理
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年5月, column 科学推理, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2.6·think
SuperCLUE 2026年5月 精确指令遵循
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年5月, column 精确指令遵循, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; read 2026-10-19 ✓xmodelKimi K2.6·think
SuperCLUE 2026年7月 幻觉控制
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年7月, column 幻觉控制, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; cost = SuperCLUE 2026年7月_成本 sheet (same 18 models/effort strings/builds as this board), CNY, tokens x list unit price; read 2026-10-19 ✓xmodelKimi K3·max
SuperCLUE 2026年7月 数学推理
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年7月, column 数学推理, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; cost = SuperCLUE 2026年7月_成本 sheet (same 18 models/effort strings/builds as this board), CNY, tokens x list unit price; read 2026-10-19 ✓xmodelKimi K3·max
SuperCLUE 2026年7月 智能体任务规划
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年7月, column 智能体任务规划, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; cost = SuperCLUE 2026年7月_成本 sheet (same 18 models/effort strings/builds as this board), CNY, tokens x list unit price; read 2026-10-19 ✓xmodelKimi K3·max
SuperCLUE 2026年7月 智能体编程
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年7月, column 智能体编程, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; cost = SuperCLUE 2026年7月_成本 sheet (same 18 models/effort strings/builds as this board), CNY, tokens x list unit price; read 2026-10-19 ✓xmodelKimi K3·max
SuperCLUE 2026年7月 科学推理
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年7月, column 科学推理, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; cost = SuperCLUE 2026年7月_成本 sheet (same 18 models/effort strings/builds as this board), CNY, tokens x list unit price; read 2026-10-19 ✓xmodelKimi K3·max
SuperCLUE 2026年7月 精确指令遵循
superclue-general
SuperCLUE general leaderboard (CLUE), edition 2026年7月, column 精确指令遵循, closed-book private questions via API, n/attempts/judge not stated on the sheet; sheet 总排行榜 only; cost = SuperCLUE 2026年7月_成本 sheet (same 18 models/effort strings/builds as this board), CNY, tokens x list unit price; read 2026-10-19 ✓xmodelKimi K3·max
SuperCLUE-Terminal (Chinese agentic terminal coding)
superclue-terminal
SuperCLUE-Terminal (CLUE team, Aug 2026; primary board/WeChat article NOT reached): Chinese-language real programming tasks of 50-110 interaction rounds each, all models through a Claude Code-based generic agent framework; numbers read from two secondary pages dated 14 Aug 2026 (we0.ai, news.aibase.com) that agree; cost = reported mean per-task cost in CNY (benchmark execution cost, not provider pricing); read 2026-10-20 ✓xmodelKimi K3·max
Surface Evolver bench
yhenon-surface-evolver
yhenon/surface-evolver-llm-eval aggregates.json · 16 runs/model, OpenRouter recorded cost, Opus 4.8 none/high, Sonnet 5 medium, Fable 5 high ✓xmodelKimi K2.7 Code·solo
surfevolver-pass
yhenon-surface-evolver
—xmodelKimi K2.7 Code·solo
Surge AI · Antidote: Everyday Edition
surgehq.ai
Surge AI Antidote: Everyday Edition leaderboard (surgehq.ai/benchmarks/antidote); Elo from human preference comparisons (95% CI in flags); effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max · Kimi K2.6·think
Surge AI · Chartography
surgehq.ai
Surge AI Chartography leaderboard (surgehq.ai/benchmarks/chartography); score in %; effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max · Kimi K2.6·think
Surge AI · ComplexConstraints
surgehq.ai
Surge AI ComplexConstraints leaderboard (surgehq.ai/benchmarks/complex-constraints); score in %; effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max · Kimi K2.6·think
Surge AI · DAYJOB: Finance
surgehq.ai
Surge AI DAYJOB: Finance leaderboard (surgehq.ai/benchmarks/dayjob-finance); score in %; effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max
Surge AI · DAYJOB: Healthcare
surgehq.ai
Surge AI DAYJOB: Healthcare leaderboard (surgehq.ai/benchmarks/dayjob-healthcare); score in %; effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max
Surge AI · EnterpriseBench: CoreCraft
surgehq.ai
Surge AI EnterpriseBench: CoreCraft leaderboard (surgehq.ai/benchmarks/enterprisebench-corecraft); score in %; effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max · Kimi K2.6·think
Surge AI · GDP.pdf
surgehq.ai
Surge AI GDP.pdf leaderboard (surgehq.ai/benchmarks/gdp-pdf); score in %; effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max · Kimi K2.6·think
Surge AI · GDP.xlsx
surgehq.ai
Surge AI GDP.xlsx leaderboard (surgehq.ai/benchmarks/gdp-xlsx); score in %; effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max
Surge AI · HANDBOOK.md Agents
surgehq.ai
Surge AI HANDBOOK.md Agents leaderboard (surgehq.ai/benchmarks/handbook); score in %; effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K2.6·think · Kimi K3·max
Surge AI · Hemingway-bench
surgehq.ai
Surge AI Hemingway-bench leaderboard (surgehq.ai/benchmarks/hemingway-bench); Elo from human preference comparisons (95% CI in flags); effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max · Kimi K2.6·think
Surge AI · Riemann-bench
surgehq.ai
Surge AI Riemann-bench leaderboard (surgehq.ai/benchmarks/riemann-bench); score in %; effort from the model label; vendor-run; no cost; read 2026-10-15 ✓xmodelKimi K3·max · Kimi K2.6·think
surge-twi-oct9
surge-twi
—xmodelKimi K3·max
SVGBench README leaderboard · score %
svgbench-github
SVGBench (github.com/johnbean393/SVGBench) README leaderboard at commit 55c1cde8 (2026-02-12): single score column, 100+ SVG-generation questions with requirement-based scoring; effort = the label in parentheses where present; read 2026-10-20 ✓xmodelKimi K2 Thinking·solo
SWE-bench Verified (GLM-4.7 card)
zhipu
SWE-bench Verified; GLM-4.7 card 2026-01-29; settings: temperature 0.7 top-p 1.0 max new tokens 16384; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
tau2-Bench (GLM-4.7 card)
zhipu
tau2-Bench; GLM-4.7 card 2026-01-29; settings: temperature 0 max new tokens 16384; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
TaxCalcBench TY25, no web search (column-tax/tax-calc-bench)
column-tax-tax-calc-bench
50 TY25 federal+state tax-return cases, 1 run each; strict-correct-return % is the score (lenient% also in the evidence but not used here per instructions); cost = mean litellm-estimate USD/return (n priced noted when <50); source: taxcalc_ty25_claude_by_effort.txt, re-verified against raw evaluation_result_*.md for opus-5.5/ultrathink (40.00%/50.00%/$1.135, exact match); read 2026-10-01 ✓xmodelKimi K3·max
TaxCalcBench TY25, with web search (column-tax/tax-calc-bench)
column-tax-tax-calc-bench
Same as taxcalc-ty25-notool but with the web-search tool enabled; read 2026-10-01 ✓xmodelKimi K3·max
Terminal Bench 2.0 (GLM-4.7 card)
zhipu
Terminal Bench 2.0; GLM-4.7 card 2026-01-29; settings: temperature 0.7 top-p 1.0 16384; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
Terminal Bench Hard (GLM-4.7 card)
zhipu
Terminal Bench Hard; GLM-4.7 card 2026-01-29; settings: temperature 0.7 top-p 1.0 16384; footnotes on judges/harness not in the card; read 2026-10-19 ✓xmodelKimi K2 Thinking·solo
Tinybird SQL benchmark
llm-benchmark.tinybird.live
Tinybird LLM SQL benchmark, 50 natural-language questions answered in SQL over 200M GH Archive rows, via OpenRouter · composite score (efficiency, exactness); no effort set (default); no cost; read 2026-10-01 ✓xmodelKimi K2.7 Code·solo
Toolathlon-Verified leaderboard (current, 108 tasks)
toolathlon
Toolathlon-Verified (released 30 Jun 2026), 108 tasks, 32 MCP servers / 604 tools + 7 local toolkits, Pass@1 % (+-CI) over 3 runs; rows carry the 'Evaluated by us' badge unless noted; Agent 'Default'; page fetched 2026-10-20 (newest row 2026-08-30); cells identical to held cells (vendor cards/blogs copying the board) dropped; read 2026-10-20 ✓xmodelKimi K3·max · Kimi K2.7 Code·solo
tracker-quality-score
github-startakovsky-llm-model-tracker
—xmodelKimi K2.7 Code·solo
UGI Leaderboard — UGI composite (all Claude models × efforts)
ugi-leaderboard
DontPlanToEnd/UGI-Leaderboard, fixed prompt set, composite UGI score; test dates vary by model (CSV Test Date column); read 2026-10-05 ✓xmodelKimi K2 Thinking·solo
UGI Leaderboard — World Model factual recall (all Claude models × efforts)
ugi-leaderboard
DontPlanToEnd/UGI-Leaderboard, recipe/geoguessr/weight/music percent-error battery; test dates vary by model; read 2026-10-05 ✓xmodelKimi K2 Thinking·solo
UGI Leaderboard — Writing quality (all Claude models × efforts)
ugi-leaderboard
DontPlanToEnd/UGI-Leaderboard, fixed prompt set, LLM-judged Writing score; test dates vary by model (CSV Test Date column); read 2026-10-05 ✓xmodelKimi K2 Thinking·solo
Vals AIME
vals.ai
Vals AI · AIME, compute_effort null (effort unverified), board 2026-04-16 ✓xmodelKimi K2 Thinking·solo
Vals CaseLaw v2
vals.ai
Vals AI · CaseLaw v2, compute_effort null (effort unverified), board 2026-05-04 ✓xmodelKimi K2 Thinking·solo
Vals Code Migration
vals.ai
Vals AI · Code Migration, 10 Claude entries, compute_effort max (Haiku thinking null) — snapshot 27 Sep, re-priced ✓xmodelKimi K2.7 Code·solo · Kimi K3·max
Vals CorpFin v2
vals.ai
Vals AI · CorpFin v2, Claude models at their stated compute_effort, measured cost/test ✓xmodelKimi K3·max
Vals CyberBench
vals.ai
Vals AI · OSS-Fuzz PoC+patch, mini-swe-agent, 5 Claude models at max — snapshot 27 Sep, re-priced ✓xmodelKimi K3·max
Vals Finance Agent v2
vals.ai
Vals AI · Finance Agent v2, Claude models at their stated compute_effort, measured cost/test — snapshot 27 Sep, re-priced ✓xmodelKimi K3·max
Vals GPQA
vals.ai
Vals AI · GPQA, Claude models at their stated compute_effort, measured cost/test, near-ceiling → cost only ✓xmodelKimi K3·max
Vals HLAB
vals.ai
Vals AI · Harvey's Legal Agent Benchmark, all at max, several at floor — snapshot 27 Sep, re-priced ✓xmodelKimi K3·max
Vals Index (29 Sep)
vals.ai
vals.ai Vals Index as recomputed on 29 Sep 2026 (older models dropped), max effort, accuracy and cost per test ✓xmodelKimi K3·max
Vals IOI
vals.ai
Vals AI · IOI, Claude models at their stated compute_effort, measured cost/test — snapshot 27 Sep, re-priced ✓xmodelKimi K3·max
Vals Legal Research
vals.ai
Vals AI · Legal Research, Claude models at their stated compute_effort, measured cost/test — snapshot 27 Sep, re-priced ✓xmodelKimi K3·max
Vals LegalBench
vals.ai
Vals AI · LegalBench, Claude models at their stated compute_effort, measured cost/test — snapshot 27 Sep, re-priced ✓xmodelKimi K3·max
Vals LiveCodeBench
vals.ai
Vals AI · LiveCodeBench, Claude models at their stated compute_effort, measured cost/test ✓xmodelKimi K2.7 Code·solo · Kimi K3·max
Vals MedCode
vals.ai
Vals AI · ICD-10-CM coding, 2,755 records, all at max ✓xmodelKimi K3·max
Vals MedQA
vals.ai
Vals AI · MedQA, Sonnet 4.6 vs Haiku 4.5 thinking, effort null, board 2026-04-16 ✓xmodelKimi K2 Thinking·solo
Vals MedScribe
vals.ai
Vals AI · MedScribe, Claude models at their stated compute_effort, measured cost/test ✓xmodelKimi K3·max
Vals MMLU-Pro
vals.ai
Vals AI · MMLU-Pro, Claude models at their stated compute_effort, measured cost/test, near-ceiling → cost only ✓xmodelKimi K3·max
Vals MMMU
vals.ai
Vals AI · MMMU, Claude models at their stated compute_effort, measured cost/test ✓xmodelKimi K3·max
Vals MortgageTax
vals.ai
Vals AI · MortgageTax, Claude models at their stated compute_effort, measured cost/test ✓xmodelKimi K3·max
Vals ProgramBench
vals.ai
Vals AI · ProgramBench, max, mostly floor scores, 3 entries without cost — snapshot 27 Sep, re-priced ✓xmodelKimi K2.7 Code·solo · Kimi K3·max
Vals ProofBench
vals.ai
Vals AI · ProofBench v1.1, all at max (Opus 5 now max in payload), 4 saturated — snapshot 28 Sep, re-priced ✓xmodelKimi K3·max
Vals Public Benefits v1.1
vals.ai
Vals AI · Public Benefits v1.1, Claude models at their stated compute_effort, measured cost/test — snapshot 27 Sep, re-priced ✓xmodelKimi K3·max
Vals SAGE
vals.ai
Vals AI · grading student math work against a rubric, all at max ✓xmodelKimi K3·max
Vals SWE-bench
vals.ai
Vals AI · SWE-bench, Claude models at their stated compute_effort, measured cost/test ✓xmodelKimi K2.7 Code·solo · Kimi K3·max
Vals TaxEval v2
vals.ai
Vals AI · TaxEval v2, Claude models at their stated compute_effort, measured cost/test ✓xmodelKimi K3·max
Vals TB 2.1
vals.ai
Vals AI · Terminal-Bench 2.1, Claude models at their stated compute_effort (Sonnet 5.5, Opus 5.5 high) — snapshot 27 Sep, re-priced ✓xmodelKimi K2.7 Code·solo
Vals Time Horizon KSP
vals.ai
Vals AI · Time Horizon Index: KSP, 3 Claude at max, board 2026-09-14 ✓xmodelKimi K3·max
vals.ai SWE-bench Multimodal (JS front-end issues with screenshots)
vals.ai
Vals-run agent; resolve rate overall, cost per test (USD); effort max per vals model pages; page updated 2026-10-13; read 2026-10-15 ✓xmodelKimi K3·max
valscodemig11
vals.ai
—xmodelKimi K3·max
valsemb11
vals.ai
—xmodelKimi K3·max
valsfab21
vals.ai
—xmodelKimi K3·max
valsfinindex
vals.ai
—xmodelKimi K3·max
valshlab-2026-10-21
vals.ai
—xmodelKimi K3·max
valsindex4
vals.ai
—xmodelKimi K3·max
valsinfprovider
vals.ai
—xmodelKimi K3·max
valslegalindex
vals.ai
—xmodelKimi K3·max
valssage11
vals.ai
—xmodelKimi K3·max
valstaxwp
vals.ai
—xmodelKimi K3·max
Vending-Bench 2 (Andon Labs)
andonlabs-vb2
andonlabs.com/evals/vending-bench-2: mean final money balance after a simulated year (start $500), mean of runs; 3 rows read from the page, the rest relayed by Epoch's mirror; no effort stated by the page; no cost; read 2026-10-15 ✓xmodelKimi K2.7 Code·solo
vlm-exam counting
github-vlmexam
roboflow/vlm-exam · 74 samples, LLM-judge accuracy ✓xmodelKimi K2.6·nothink
vlm-exam detection
github-vlmexam
roboflow/vlm-exam · 250 images, mAP@50 ✓xmodelKimi K2.6·nothink
vlm-exam extraction
github-vlmexam
roboflow/vlm-exam · 97 samples, LLM-judge accuracy; same models as OCR ✓xmodelKimi K2.6·nothink
vlm-exam identification
github-vlmexam
roboflow/vlm-exam · 32 samples, near ceiling ✓xmodelKimi K2.6·nothink
vlm-exam OCR
github-vlmexam
roboflow/vlm-exam · 37 samples, similarity; Fable 5.1 & Opus 5.5 low/high ×3 runs, others low ×1 ✓xmodelKimi K2.6·nothink
vlmexam-text
github-vlmexam
—xmodelKimi K2.6·nothink · Kimi K3·high
vulcanbench3
github-vulcanbench
—xmodelKimi K3·max
τ^τ-Bench
arxiv-2609.04611
arXiv 2609.04611 · 53 agent-construction tasks in Claude Code, Opus 5 vs Sonnet 5 both at max, build spend at list ✓xmodelKimi K3·max
How the numbers are built — the method

Every value in this report is computed from the raw measurements by one procedure — nothing is hand-entered. The full methodology is in METHODOLOGY.md (from collection to the fit) and DISPLAY-METHODOLOGY.md (from the fit to the tier picks); the validation scripts are in model/validation.

In short : scores from different benchmarks do not share a scale — a benchmark near its ceiling compresses differences, a hard one inflates them — so the report never averages raw scores or score ratios. A latent-quality model estimates, for every benchmark, its own zero and its own gain, and for every (model, effort) couple one quality and one cost. Each couple stays atomic: no model×effort separability is assumed.

The procedure

  1. Every number traced to its source. Each source is recorded as facts: the value as the source writes it, the exact text or table cell it is read from, the URL and the date of the capture; agents read every value line by line, and a measurement whose facts lack their proof stays out. Registries of models (the names each was served under, its effort settings and defaults, dated), routes (makers' APIs, resellers, routers, harnesses), benchmarks and their versions, publishers and dated prices let one build decide, by a fixed decision tree, which model, configuration and price each measurement belongs to.
  2. Same task, same configuration. Measurements are compared only within a group — one source, one task, one harness, one setting, one scale, one version of a board; a group that mixes them is split. Every run is dated, and its model and effort are read as the maker and every route in between served them on that date — never assumed: a route the source does not name is never presumed to be the maker's API, and every route that could have served the run must give the same configuration. At the least doubt on the model, the effort or what was run, the measurement stays in the data file and out of the fit.
  3. Costs at today's prices. Every cost is put on the model's current price from the prices in force on its run date: the source's own figure at that price, the price change applied to the run's own token counts, or, when the token mix is unknown, a price band whose width is added to the cost's precision. A cost that cannot be brought to the current price stays out.
  4. Each metric on its natural scale. A bounded score (a percentage, a score out of N) goes through a logit, which removes floor and ceiling compression; money and points go through a log; an Elo or an unknown composite is left as is; cost goes through a log. The scale is read from the metric's label, never from the benchmark's identity; whether lower is better is read from the benchmark's registry, and such a score is turned in its own unit first (an attack success rate of 30 % reads 70 %); a score whose direction is not established stays out. On the logit scale a score near 0 % or 100 % is noisier, and weighs accordingly. Every value carries its reading precision: the rounding of a printed number, and for a value read off a chart, its resolution and the chart's own error (measured on runs drawn on two charts, about 2 % on a digitised cost).
  5. One model for all groups. Every group gets its own offset, gain and noise, so the unit and zero of a metric never matter. A couple's latent value is shared by all groups; on top of it, a publisher's systematic effect on a couple and on a model (shared by the model's effort levels) and a couple's deviation on a task type are estimated, so a publisher with many benchmarks, or a couple tested mostly on one kind of task, cannot tilt the result. The sizes of these effects are estimated for each model maker on its own, not tied to the other makers'; a maker's own charts about its own models carry an estimated home effect, also each maker's own, left out of every value. The fit takes the couples of eight makers measured in the same groups (Anthropic, OpenAI, Google, xAI, DeepSeek, Qwen, GLM, Kimi), on one scale; this page shows the Kimi couples. Robust (Student-t) noise whose tails are estimated; early-access runs are down-weighted by an estimated factor. No couple serves as a reference in the fit: the latent values sum to zero. Fitted in Stan, published only if the standard convergence checks pass.
  6. Weights follow from the data. A group of k couples informs the spacing of couples in proportion to its discrimination² × (k − 2); a saturated or noisy benchmark weighs little; a publisher's weight saturates at its own systematic effect. No weight is set per source; the one weight set by rule is that of a single run: when a source gives k quality measurements of one run (several metrics, sub-scores or slices of the same executions), each weighs 1/k, so the run counts as one measurement.
  7. Nothing counts twice. One execution printed twice — a system card reprinting an earlier card's series, the same run drawn on two charts, a board copying another's numbers, one run scored with two metrics — is detected and counted once: its score and its cost must both agree, within their reading precisions and a rounding made the wrong way. Versions of one board that re-grade the same runs (recognised by identical costs) count those runs once. Before the fit, two groups that still share identical costs on three couples or more stop it until the copy is resolved or the coincidence proven from the sources. An index keeps, for a couple whose index components are already in the data, only what its missing components add, or is left out.
  8. Values and intervals. Quality = a couple's latent quality θ, shown as the score it predicts averaged over the benchmark panel; cost = its cost on a task of typical size, shown as a multiple of the cheapest couple. No couple serves as a reference, in the fit or on the page. Each couple carries its own 16–84 % interval, so two couples compare through their two intervals. A couple measured by a single publisher is not shown.
  9. From the fit to the picks. Every decision is taken on the latent scale (log cost, θ). The Pareto frontier keeps the couples nothing beats on both axes; they are the candidates of every pick. A price trend — what a given quality costs today, fitted on the latest model of each family — says how many times cheaper or dearer each couple is than the going rate. The four tiers spread their targets across the frontier and pick, around each target, the couple that brings the most quality for its cost, each tier paying for quality from a quarter of the market price (grunt work) to four times it (cutting-edge thinking); the crown goes to the couple furthest below the trend.

Key values

  • Set by hand, never tuned to the data: weakly informative priors (half-Student-t(3, 0, 2.5) on every spread, Gamma(2, 0.1) on the noise tails); a logit kept half a step inside 0 and 100 %; a printed value's precision at its rounding (last digit / √12); an effect seen in a single measurement left out; a couple shown only when two publishers measured it; a reprint recognised by values that agree within their precisions, on two values at least and on all they share but one.
  • Uncertainty. Every interval is 16–84 % (one standard deviation): the couple's own uncertainty.
  • Published only when the fit is sound: R̂ ≤ 1.01 and at least 400 effective draws on every parameter, no divergent transition. The Monte Carlo error of each value is checked against its last displayed digit; in this fit one figure on this page can move by one unit in that digit (the cost of Kimi K3 at high), far inside their intervals.
  • Display choices: the quality axis is compressed below the weakest tier target; the frontier, the price trend and the four tiers use constants listed with their reasons in DISPLAY-METHODOLOGY.md.

Techniques and software

  • Bayesian hierarchical model sampled by Hamiltonian Monte Carlo with the No-U-Turn sampler (Hoffman & Gelman 2014), written in Stan and run with nutpie on both axes, six chains whose draws continue in the same run until the criteria are met (CmdStan only to resume a stopped fit from its stored draws, or to restart a chain left in a far region from a good chain's last draw); built and checked stage by stage following the Bayesian workflow.
  • Convergence checked with the rank-normalised R̂ and effective sample sizes of Vehtari et al. 2021, computed with ArviZ.
  • Intervals as quasi-standard errors (Firth & de Menezes 2004), so that any two couples compare through their own intervals.
  • Scales: the empirical logit of bounded scores, with the binomial shape of its noise.

Checks (on the fit of 29 September 2026, before the other makers' couples joined the data). Scores held out of the fit and predicted: median error 1.96 points, against 4.78 for averaging score ratios; the 16–84 % predictive interval covers 66 % of them. On synthetic data with known qualities, pairs of couples in the wrong order: 0.1–0.8 % (0.4–2.2 % for score ratios). Without the largest publisher, the model vendor itself, costs move by 2.4 % and qualities by 0.9 % at the median. Details and scripts: METHODOLOGY.md.

Alexandre Gensse · github.com/alexandregensse-blip

All figures are indicative and for informational purposes only — derived from public third-party measurements, not an official benchmark, and not affiliated with or endorsed by any model maker. Verify before relying on any value.

Data and text under CC BY 4.0 · code under MIT · download the data (CSV)