Benchmark Registry

Moonshot AI

Benchmark results for 3 Moonshot AI models from official publications. Updated 2026-10-07.

Established
Not available
Latest model
Kimi K3 (July 2026)

Benchmarks

119 results

ModelBenchmarkScoreSourceRegistry No.
Kimi K2.5 (Thinking on)Chartography October 2026 leaderboard12.6%surgehq.aiArchived copyChartography Pass@1Reported Benchmark leaderboardSurge AIIndependentRecord120001
Kimi K2.5SWE Atlas Refactoring — mini-swe-agent21.0%labs.scale.comArchived copyMean pass@1Reported Benchmark leaderboardScale AIIndependentRecord120001
Kimi K2.5SWE Atlas Codebase QnA — Public; mini-swe-agent13.1%labs.scale.comArchived copyMean pass@1Reported Benchmark leaderboardScale AIIndependentRecord120001
Kimi K2.5GPQA Diamond — Epoch AI harness87.6%epoch.aiArchived copyGPQA Diamond AccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120001
Kimi K2.5SimpleQA Verified Original — Epoch AI harness34.3%epoch.aiArchived copyAccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120001
Kimi K2.5 (Thinking on)Riemann-bench October 2026 leaderboard12.0%surgehq.aiArchived copyRiemann-bench Pass@1Reported Benchmark leaderboardSurge AIIndependentRecord120001
Kimi K2.5SWE-bench Verified — Epoch AI harness73.8%epoch.aiArchived copyResolved TasksReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120001
Kimi K2.6AutomationBench 1.0.62.6%zapier.comArchived copyTask pass rateReported Benchmark leaderboardZapierIndependentZapier AutomationBench TeamRecord120002
Kimi K2.6GPQA Diamond — Epoch AI harness90.8%epoch.aiArchived copyGPQA Diamond AccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120002
Kimi K2.6 (Thinking on)Riemann-bench October 2026 leaderboard7.2%surgehq.aiArchived copyRiemann-bench Pass@1Reported Benchmark leaderboardSurge AIIndependentRecord120002
Kimi K2.6SimpleQA Verified Original — Epoch AI harness34.9%epoch.aiArchived copyAccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120002
Kimi K2.6SWE-bench Verified — Epoch AI harness76.7%epoch.aiArchived copyResolved TasksReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120002
Kimi K2.6 (Thinking on)Chartography October 2026 leaderboard12.2%surgehq.aiArchived copyChartography Pass@1Reported Benchmark leaderboardSurge AIIndependentRecord120002
Kimi K3 (max)CharXiv Descriptive and reasoning — Mercor single-shot93.9%mercor.comArchived copyCharXiv AccuracyReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)MMMU-Pro 1,730 questions — Mercor single-shot80.7%mercor.comArchived copyMMMU-Pro accuracySetting: Not statedReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)DeepSWE 1.1 — Mercor mini-swe-agent, 500 steps66.4%mercor.comArchived copyDeepSWE Pass@1Reported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (Max)Riemann-bench October 2026 leaderboard37.6%surgehq.aiArchived copyRiemann-bench Pass@1Reported Benchmark leaderboardSurge AIIndependentRecord120003
Kimi K3 (max)Terminal-Bench 4.0 — Mercor mini-swe-agent26.3%mercor.comArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3AutomationBench 1.0.622.7%zapier.comArchived copyTask pass rateReported Benchmark leaderboardZapierIndependentZapier AutomationBench TeamRecord120003
Kimi K3 (Max)Chartography October 2026 leaderboard26.6%surgehq.aiArchived copyChartography Pass@1Reported Benchmark leaderboardSurge AIIndependentRecord120003
Kimi K3FrontierSWE 2 — October 2026 evaluation25.9%frontierswe.comArchived copyFrontierSWE ScoreReported Benchmark leaderboardProximalIndependentRecord120003
Kimi K3 (max)MedXpertQA MM — Mercor single-shot73.2%mercor.comArchived copyAccuracyReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)SWE-bench Verified — Mercor mini-swe-agent, 1000 steps83.3%mercor.comArchived copyResolved TasksReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)APEX-Agents Original50.6%mercor.comArchived copyPass rateReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)SWE-bench Multilingual — Mercor mini-swe-agent, 1000 steps91.4%mercor.comArchived copyResolved TasksReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (high)GPQA Diamond — Epoch AI harness91.9%epoch.aiArchived copyGPQA Diamond AccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120003
Kimi K3 (max)Terminal-Bench 2.1 — Mercor Terminus 282.0%mercor.comArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)GPQA Diamond — Epoch AI harness93.1%epoch.aiArchived copyGPQA Diamond AccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120003
Kimi K3 (max)BrowseComp 130-question subset — Mercor web research agent89.0%mercor.comArchived copyBrowseComp AccuracyReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)SimpleQA Verified Original — Epoch AI harness50.6%epoch.aiArchived copyAccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120003
Kimi K3 (low)GPQA Diamond — Epoch AI harness84.8%epoch.aiArchived copyGPQA Diamond AccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord120003
Kimi K3 (max)GPQA Diamond — Mercor single-shot92.7%mercor.comArchived copyGPQA Diamond AccuracyReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)MMLU-Pro Original — Mercor single-shot89.3%mercor.comArchived copyAccuracyReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)Humanity's Last Exam Text only — Mercor single-shot46.5%mercor.comArchived copyHumanity's Last Exam AccuracyReported Benchmark leaderboardMercorIndependentRecord120003
Kimi K3 (max)SWE Atlas Codebase QnA — Mercor mini-swe-agent, 250 steps52.7%mercor.comArchived copyMean pass@1Reported Benchmark leaderboardMercorIndependentRecord120003
Kimi K2.5 (thinking)Vibe Code Bench 1.1 — OpenHands, October 2026 evaluation17.5%vals.aiArchived copyVibe Code Bench ScoreReported Benchmark leaderboardVals AIIndependentRecord120001
Kimi K2.5 (thinking)Finance Agent Benchmark 2 — October 2026 evaluation35.8%vals.aiArchived copyFinance Agent AccuracyReported Benchmark leaderboardVals AIIndependentRecord120001
Kimi K2.5 (thinking)HLAB October 2026 evaluation0.0%vals.aiArchived copyHLAB Task Pass RateReported Benchmark leaderboardVals AIIndependentRecord120001
Kimi K2.6Finance Agent Benchmark 2 — October 2026 evaluation44.9%vals.aiArchived copyFinance Agent AccuracyReported Benchmark leaderboardVals AIIndependentRecord120002
Kimi K2.6Vibe Code Bench 1.1 — OpenHands, October 2026 evaluation37.9%vals.aiArchived copyVibe Code Bench ScoreReported Benchmark leaderboardVals AIIndependentRecord120002
Kimi K2.6HLAB October 2026 evaluation1.7%vals.aiArchived copyHLAB Task Pass RateReported Benchmark leaderboardVals AIIndependentRecord120002
Kimi K3 (max)Terminal-Bench 4.0 — Vals AI mini-swe-agent17.2%vals.aiArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardVals AIIndependentRecord120003
Kimi K3 (max)Vals Index 2.150.3%vals.aiArchived copyVals Index GDP-weighted AccuracyReported Benchmark leaderboardVals AIIndependentRecord120003
Kimi K3 (max)HLAB October 2026 evaluation12.9%vals.aiArchived copyHLAB Task Pass RateReported Benchmark leaderboardVals AIIndependentRecord120003
Kimi K3 (max)Finance Agent Benchmark 2 — October 2026 evaluation53.1%vals.aiArchived copyFinance Agent AccuracyReported Benchmark leaderboardVals AIIndependentRecord120003
Kimi K3 (max)Vibe Code Bench 1.1 — OpenHands, October 2026 evaluation85.0%vals.aiArchived copyVibe Code Bench ScoreReported Benchmark leaderboardVals AIIndependentRecord120003
Kimi K3PostTrainBench 1.2 — Claude Code31.5%posttrainbench.comArchived copyPostTrainBench Weighted AggregateReported Benchmark leaderboardPostTrainBenchIndependentPostTrainBench TeamRecord120003
Kimi K3PostTrainBench 1.1 — Claude Code32.0%posttrainbench.comArchived copyPostTrainBench Weighted AggregateReported Benchmark leaderboardPostTrainBenchIndependentPostTrainBench TeamRecord120003
Kimi K2.5Humanity's Last Exam Full set — Scale AI evaluation24.4%labs.scale.comArchived copyHumanity's Last Exam AccuracyReported Benchmark leaderboardScale AIIndependentRecord120001
Kimi K2.5 (thinking)Terminal-Bench 2.1 — Vals AI Terminus 241.9%vals.aiArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardVals AIIndependentRecord120001

Moonshot AI models