RankingEffort benchmarks

Kimi K3 · High

Compare models

81 measured results · 5 effort settings · 56 benchmark/harness combinations

Model evidence and sources →

Change model

Model

Current configurationSources

Compare with

Choose a second model and its effort to see both profiles and benchmark differences below.

Compare withModel and configuration

Capability profile

One configuration, seven capabilities

Comparison score · 0–100

AKimi K3 · High1 of 7 capabilities supported

Gaps are unknown, not zero. Hollow points are preliminary.

Agentic
Unknown
Hard reasoning
A41.5Supported
Coding
Unknown
Human pref
Unknown
Knowledge
Unknown
Multimodal
Unknown
Long context
Unknown

Each profile belongs to the named configuration. Capability scores use a shared panel for all axes, distinct from the overall ranking panel; they are not accuracy percentages or direct head-to-head wins.

Capability score evidence

Kimi K3 · High: 36.5 overall comparison score. Meets the overall evidence requirements.

Hard reasoning only. The available task mix affects this estimate; missing areas remain unknown.

Which sources carry the weight?

ARC Prize · 70.0% direct comparison weight · independent of the model provider
Epoch AI · 30.0% direct comparison weight · independent of the model provider

Which benchmark families carry the weight?

ARC-AGI · 70.0% direct comparison weight
GPQA · 15.1% direct comparison weight
Aime · 14.9% direct comparison weight

These shares describe direct comparison weight, not fractions of the final score or confidence. Source balancing does not give every family equal weight or remove correlations between overlapping sources.

How sensitive is this score?

The audit removes one source at a time and refits against the same overall reference panel. These published results use the matching dataset audited 2026-10-23.

Both ranks use the same surviving configurations for each removal. The cohort starts with 444 configurations supported in the original audit, including separate effort settings. These are diagnostic positions, not the default leaderboard ranks. Eligibility is not rechecked after removal.

What changes when a source is removed?
Removed source	Rank before → after	Rank movement	Cohort remaining / lost support	Score after (change)
Artificial Analysis	88 → 91	Down 3	203 remaining / 241 lost	34.6 (-2.0 points)
ARC Prize	97 → 69	Up 28	444 remaining / 0 lost	52.2 (+15.7 points)
Datacurve	97 → 97	No movement	444 remaining / 0 lost	31.1 (-5.4 points)
LiveCodeBench	97 → 96	Up 1	444 remaining / 0 lost	36.7 (+0.1 points)
Scale	97 → 96	Up 1	444 remaining / 0 lost	38.5 (+2.0 points)
Terminal-Bench	97 → 100	Down 3	444 remaining / 0 lost	36.3 (-0.2 points)
Cognition	97 → 99	Down 2	444 remaining / 0 lost	35.8 (-0.7 points)
LiveBench	97 → 96	Up 1	443 remaining / 1 lost	36.7 (+0.1 points)
SWE-rebench	97 → 96	Up 1	444 remaining / 0 lost	36.2 (-0.3 points)
Epoch AI	96 → 104	Down 8	431 remaining / 13 lost	34.7 (-1.9 points)
Vals AI	97 → 100	Down 3	443 remaining / 1 lost	35.1 (-1.4 points)
Provider reports · Anthropic	97 → 97	No movement	444 remaining / 0 lost	36.5 (-0.0 points)
Provider reports · Moonshot AI	97 → 97	No movement	444 remaining / 0 lost	36.1 (-0.5 points)
Provider reports · Meta	97 → 97	No movement	444 remaining / 0 lost	36.5 (-0.0 points)
Provider reports · Mistral	97 → 97	No movement	444 remaining / 0 lost	36.5 (+0.0 points)
Provider reports · OpenAI	97 → 96	Up 1	444 remaining / 0 lost	36.7 (+0.2 points)
All provider reports	97 → 96	Up 1	444 remaining / 0 lost	36.2 (-0.3 points)

Lost support means a configuration no longer has a comparison path to the complete reference panel. Its rank is unavailable. A surviving path alone does not establish leaderboard eligibility.

Overlapping sources may remain correlated even with source balancing. Leave-one-source-out movement is a sensitivity diagnostic, not a spread of source-specific ranks or a confidence measure. Scoring methodology · Dated configuration audit

Every benchmark family

Removed evidence	Recomputed score	Change
arc-agi	51.3	+14.8
deepswe	31.1	-5.4
epoch-game-puzzles	35.2	-1.3
mmmu-pro	37.8	+1.3
hle	37.7	+1.2
gpqa	35.5	-1.1
aime	37.5	+1.0
simpleqa	37.3	+0.8
frontiercode	35.8	-0.7
arc-agi-3	35.8	-0.7
mmlu-pro	37.1	+0.5
livebench-instructions	37.0	+0.4
multi-swe-bench	36.2	-0.4
swe-rebench	36.2	-0.3
swe-atlas-qna	36.3	-0.3
harvey	36.3	-0.2
livebench-language	36.8	+0.2
omniscience	36.3	-0.2
livebench-coding	36.3	-0.2
enigma-eval	36.3	-0.2
ifbench	36.4	-0.2
briefcase	36.4	-0.1
tau-bench	36.4	-0.1
apex-agents	36.7	+0.1
automationbench	36.4	-0.1
livecodebench	36.7	+0.1
terminal-bench	36.4	-0.1
swe-atlas-test-writing	36.4	-0.1
vibe-code	36.4	-0.1
gdp-pdf	36.7	+0.1
livebench-reasoning	36.4	-0.1
aa-lcr	36.5	-0.1
terminal-bench-science	36.6	+0.1
analyst-agent	36.6	+0.1
gdpval	36.5	-0.1
vals-legal-research	36.5	-0.1
frontiermath	36.6	+0.1
vals-code-migration	36.5	-0.1
vals-finance-agent	36.5	-0.1
itbench	36.5	-0.1
vals-tax-agent	36.5	-0.1
critpt	36.5	-0.0
swe-atlas-refactoring	36.5	-0.0
vals-excel-modeling	36.5	-0.0
livebench-math	36.6	+0.0
enterprise-ops	36.5	-0.0
gmmlu	36.6	+0.0
livebench-data	36.5	-0.0
math500	36.5	+0.0
swe-bench-pro	36.5	+0.0

What evidence is missing from the fit?

Across all collected settings for this model: 183 observations, 70 contributing, 0 matched without graph weight and 113 excluded. Counts do not establish rank eligibility.

7 · Benchmark family or source protocol has not been reviewed
2 · Original SciCode grading defects are under review; retained as supporting evidence, excluded from aggregate and capability scores. Corrected SciCode-Verified requires a separate protocol review. Sources: https://arxiv.org/abs/2608.04975 and https://artificialanalysis.ai/evaluations/scicode
10 · Evaluator has not been reviewed for independent admission
2 · The evaluator reports no named effort control. This does not identify a non-reasoning configuration.
4 · Effort is not identified
1 · The task table records 165 trials of the nominal 170.
17 · No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately
5 · Narrow specialist biological, medical or security evaluation; preserved as raw evidence outside general capability scoring.
45 · Supporting evidence outside the reviewed capability core
7 · Kimi K3 Evaluation Details cites Artificial Analysis; retain as published context, not a new Moonshot comparison.
2 · No matched opponent in this evaluation unit
1 · Historical provider claim; secondary and superseded for current FrontierSWE comparisons by the independent Proximal FrontierSWE v2 mean@5 board at https://www.frontierswe.com/. Version-unspecified claims are not v2 results. Excluded from weighted Overall.
1 · Kimi K3 Evaluation Details cites official Agents Last Exam leaderboard; retain as published context, not a new Moonshot comparison.
1 · Kimi K3 Evaluation Details cites Artificial Analysis / APEX-Agents leaderboard; retain as published context, not a new Moonshot comparison.
3 · Kimi K3 Evaluation Details cites Vals AI; retain as published context, not a new Moonshot comparison.
1 · The GLM card attributes this evaluation to Artificial Analysis; it is not a new Z.ai comparison.
1 · Composite index or publisher-specific internal evaluation lacks a stable general-capability comparison unit. Composite index excluded to avoid double counting its ingredients.
3 · Legacy self-report lacks a reviewed comparison configuration

Inspect every observation and exclusion →

Benchmark results

These are published benchmark scores. Supported effort estimates also appear on the Capability leaderboard. Results from different harnesses occupy different rows. Missing cells stay unknown; no scores are copied between effort levels.

Benchmark / source	Low	High	Max	Default	Unspecified
AA-AnalystAgentArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Share of tasks solved on all 5 of 5 attempts · 80 tasks, 5 attempts per task · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	38.75% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
AA-Briefcase EloArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details AA-Briefcase v1.1 is an agentic knowledge work benchmark developed by Artificial Analysis. AA-Briefcase Elo is a combined metric that aggregates rubric pass rate, analytical quality Elo and presentation Elo · Higher is better · Evaluation results measured independently by Artificial Analysis Unit: elo. Source collected 2026-10-18.	985.76 Elo Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	1,499.8 Elo Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
APEX-Agents-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	41.3% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
AA-LCR v1.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	79.33% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	88.67% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
AutomationBench-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Share of task objectives completed with no guardrail violations · Higher is better · Benchmark developed by Zapier · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	49.26% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	58.27% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
CritPtArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Benchmark developed by Argonne and UIUC, with contributions from 60+ researchers globally. CritPt is under review by its authors following external feedback · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	3.14% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	23.43% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
EnterpriseOps-Gym-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Overall task success rate across all eight enterprise domains (oracle tool mode) · Benchmark developed by ServiceNow Research · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	45.33% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
GDP.pdf: All-pass RateArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Share of attempts where every atomic criterion passed · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	18.6% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	22% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
GDPval-AA v2.1 LeaderboardArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Elo rating for performance on real-world work tasks · Anchored to DeepSeek V4.1 Flash (max) at 1600 · Higher is better · Evaluation results measured independently by Artificial Analysis Unit: elo. Source collected 2026-10-18.	1,125.75 Elo Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	1,533.13 Elo Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
GPQA DiamondArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	84.24% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	93.54% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
Harvey LAB-AA v1.1: Hallucination-Gated All-Pass RateArtificial Analysis · Stirrup; Harvey private dataset v1.1.0; 120 tasks across 24 practice areas; up to 200 turns; internet disabled; GPT-6 Sol, Grok 4.7 and Claude Opus 5.5 rubric judges; GPT-6 Sol High two-pass hallucination check; hallucination-gated all-pass averaged across judges Benchmark details Average over tasks of the share of the judge panel finding every rubric criterion passed, with a task zeroed on any material hallucination · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	5.28% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
Humanity's Last ExamArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	24.98% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	46.9% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
ITBench-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Combined public and private Site Reliability Engineering (SRE) tasks · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	47.69% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
MLCR-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Share of tasks judged accurate, complete, and concise · Expert + compound (hard) sets · Benchmark developed by Wisedocs · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	38.33% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
MMMU-ProArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	78.5% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	80.52% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
AA-Omniscience IndexArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details AA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct. · Evaluation results measured independently by Artificial Analysis Unit: index. Source collected 2026-10-18.	3.92 Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	19.7 Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
AA-Omniscience AccuracyArtificial Analysis · 6000 questions; 1 attempt; correct-answer share; GPT-5.6 Luna medium grader Benchmark details AA-Omniscience Accuracy (higher is better) measures the proportion of correctly answered questions out of all questions, regardless of whether the model chooses to answer · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	45.75% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	47.58% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
SciCodeArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Benchmark curated by scientists across 16 disciplines. SciCode is under review following an independent audit of the dataset · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	52.66% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	59.49% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
𝜏³-BankingArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Benchmark developed by Sierra Research · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	41.65% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	45.98% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
Terminal-Bench v2.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings Benchmark details Benchmark developed by the Laude Institute and open-source contributors · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	82.4% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	85.02% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
Terminal-Bench v4.0Artificial Analysis · mini-SWE-agent v2.4.6; 66 tasks; pass@1 averaged over 3 repeats; 500 steps; task timeout up to 8 hours Benchmark details Benchmark developed by the Laude Institute, Stanford, and open-source contributors · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-10-18.	12.63% Reported settings Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low	Not reported	12.63% Reported settings Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3	Not reported	Not reported
ARC-AGI-2ARC Prize · v2 Semi-Private Benchmark details Original public board. Configuration and task coverage may differ; inspect source. Unit: percent. Source collected 2026-10-18.	12.36% Reported settings Kimi K3 (Low); Source model ID: moonshot-kimi-k3-low; 120 tasks	55% Reported settings Kimi K3 (High); Source model ID: moonshot-kimi-k3-high; 120 tasks	60.42% Reported settings Kimi K3 (Max); Source model ID: moonshot-kimi-k3-max; 120 tasks	Not reported	Not reported
DeepSWE v1.1Datacurve · mini-swe-agent Benchmark details Original public board. Configuration and task coverage may differ; inspect source. Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	68.5% Reported settings kimi-k3; Reported effort: max; 113 unique tasks; 451 attempts; 4 runs	Not reported	Not reported
VulcanBench v3 · Report 08VulcanBench · VulcanBench bare-bones API loop, deterministic hidden tests in Docker Benchmark details Frozen v3, 23 post-cutoff repository tasks. Published pass@1; repeat counts and time budgets vary by report. Display only pending comparability review. Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	73.9% Reported settings Kimi K3; source label: Kimi K3 (max, 30-min budget); max effort; 23 tasks; evaluation 2026-07-19. See source for repeat counts and wall-clock budget.	Not reported	Not reported
FrontierCode 1.1 · extended-mini-swe-agentCognition · mini-swe-agent Benchmark details FrontierCode 1.1; mini-swe-agent. Open each result for its exact effort, task subset, metric and protocol details. Unit: percent. Source collected 2026-10-18.	Not reported	Not reported	Not reported	Not reported	58.19% Reported settings FrontierCode 1.1; extended; 150 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0014 The evaluator reports no named effort control. This does not identify a non-reasoning configuration. Source label needs review

What these comparisons mean

Each source reports its own effort labels. Matching labels do not establish equal compute budgets or identical fallback behavior. Open Reported settings to inspect the exact model name, agent and task coverage. Multiple results at one level remain visible rather than selecting the best. Source-reported None or Non-reasoning labels are not treated as evidence that an API supports those settings.

These measurements use their own dated source collection. The leaderboard fits identified configurations jointly; it contains no pooled model ratings. Explicit thinking budgets are separate configurations. Read the methodology.

Download all effort benchmark results · 8,916 results across 259 models · Collected 2026-10-18

Kimi K3 · High

Compare with

One configuration, seven capabilities

Agentic

Hard reasoning

Coding

Human pref

Knowledge

Multimodal

Long context

Capability score evidence

Benchmark results