Benchmark Registry

MiniMax

Benchmark results for 4 MiniMax models from official publications. Updated 2026-10-07.

Established
Not available
Latest model
MiniMax M3 (June 2026)

Benchmarks

72 results

ModelBenchmarkScoreSourceRegistry No.
MiniMax M2.5SWE Atlas Refactoring — mini-swe-agent19.5%labs.scale.comArchived copyMean pass@1Reported Benchmark leaderboardScale AIIndependentRecord140001
MiniMax M2.5SWE Atlas Codebase QnA — Public; mini-swe-agent10.3%labs.scale.comArchived copyMean pass@1Reported Benchmark leaderboardScale AIIndependentRecord140001
MiniMax M2.7 (high)SWE-bench Verified — Mercor mini-swe-agent, 1000 steps75.3%mercor.comArchived copyResolved TasksReported Benchmark leaderboardMercorIndependentRecord140002
MiniMax M2.7 (high)SWE-bench Multilingual — Mercor mini-swe-agent, 1000 steps60.8%mercor.comArchived copyResolved TasksReported Benchmark leaderboardMercorIndependentRecord140002
MiniMax M2.7 (high)GPQA Diamond — Mercor single-shot86.0%mercor.comArchived copyGPQA Diamond AccuracyReported Benchmark leaderboardMercorIndependentRecord140002
MiniMax M2.7 (high)BrowseComp 130-question subset — Mercor web research agent38.3%mercor.comArchived copyBrowseComp AccuracyReported Benchmark leaderboardMercorIndependentRecord140002
MiniMax M2.7 (high)DeepSWE 1.1 — Mercor mini-swe-agent, 500 steps0.0%mercor.comArchived copyDeepSWE Pass@1Reported Benchmark leaderboardMercorIndependentRecord140002
MiniMax M2.7 (high)Humanity's Last Exam Text only — Mercor single-shot27.1%mercor.comArchived copyHumanity's Last Exam AccuracyReported Benchmark leaderboardMercorIndependentRecord140002
MiniMax M2.7 (high)MMLU-Pro Original — Mercor single-shot79.6%mercor.comArchived copyAccuracyReported Benchmark leaderboardMercorIndependentRecord140002
MiniMax M2.7 (high)Terminal-Bench 2.1 — Mercor Terminus 254.3%mercor.comArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardMercorIndependentRecord140002
MiniMax M2.7 (high)AutomationBench 1.0.60.5%zapier.comArchived copyTask pass rateReported Benchmark leaderboardZapierIndependentZapier AutomationBench TeamRecord140002
MiniMax M3 (high)SWE-bench Verified — Mercor mini-swe-agent, 1000 steps75.5%mercor.comArchived copyResolved TasksReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)MMLU-Pro Original — Mercor single-shot84.6%mercor.comArchived copyAccuracyReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)DeepSWE 1.1 — Mercor mini-swe-agent, 500 steps9.1%mercor.comArchived copyDeepSWE Pass@1Reported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3GPQA Diamond — Epoch AI harness90.9%epoch.aiArchived copyGPQA Diamond AccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord140003
MiniMax M3 (high)APEX-Agents Original37.7%mercor.comArchived copyPass rateReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)SWE-bench Multilingual — Mercor mini-swe-agent, 1000 steps76.0%mercor.comArchived copyResolved TasksReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)Terminal-Bench 4.0 — Mercor mini-swe-agent1.0%mercor.comArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)SWE Atlas Codebase QnA — Mercor mini-swe-agent, 250 steps37.9%mercor.comArchived copyMean pass@1Reported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)Terminal-Bench 2.1 — Mercor Terminus 261.4%mercor.comArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (none)GPQA Diamond — Epoch AI harness81.3%epoch.aiArchived copyGPQA Diamond AccuracyReported Evaluated Benchmark leaderboardEpoch AIIndependentRecord140003
MiniMax M3 (high)MMMU-Pro 1,730 questions — Mercor single-shot78.9%mercor.comArchived copyMMMU-Pro accuracySetting: Not statedReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)Humanity's Last Exam Text only — Mercor single-shot37.8%mercor.comArchived copyHumanity's Last Exam AccuracyReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)MedXpertQA MM — Mercor single-shot59.7%mercor.comArchived copyAccuracyReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)GPQA Diamond — Mercor single-shot90.5%mercor.comArchived copyGPQA Diamond AccuracyReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (max)AutomationBench 1.0.64.9%zapier.comArchived copyTask pass rateReported Benchmark leaderboardZapierIndependentZapier AutomationBench TeamRecord140003
MiniMax M3 (high)BrowseComp 130-question subset — Mercor web research agent53.3%mercor.comArchived copyBrowseComp AccuracyReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M3 (high)CharXiv Descriptive and reasoning — Mercor single-shot85.9%mercor.comArchived copyCharXiv AccuracyReported Benchmark leaderboardMercorIndependentRecord140003
MiniMax M2.5Vibe Code Bench 1.1 — OpenHands, October 2026 evaluation14.9%vals.aiArchived copyVibe Code Bench ScoreReported Benchmark leaderboardVals AIIndependentRecord140001
MiniMax M2.7Finance Agent Benchmark 2 — October 2026 evaluation27.9%vals.aiArchived copyFinance Agent AccuracyReported Benchmark leaderboardVals AIIndependentRecord140002
MiniMax M2.7Vibe Code Bench 1.1 — OpenHands, October 2026 evaluation11.9%vals.aiArchived copyVibe Code Bench ScoreReported Benchmark leaderboardVals AIIndependentRecord140002
MiniMax M2.7HLAB October 2026 evaluation0.0%vals.aiArchived copyHLAB Task Pass RateReported Benchmark leaderboardVals AIIndependentRecord140002
MiniMax M3Vals Index 2.136.5%vals.aiArchived copyVals Index GDP-weighted AccuracyReported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M3Finance Agent Benchmark 2 — October 2026 evaluation48.3%vals.aiArchived copyFinance Agent AccuracyReported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M3Vibe Code Bench 1.1 — OpenHands, October 2026 evaluation47.6%vals.aiArchived copyVibe Code Bench ScoreReported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M3HLAB October 2026 evaluation4.2%vals.aiArchived copyHLAB Task Pass RateReported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M3Terminal-Bench 4.0 — Vals AI mini-swe-agent1.0%vals.aiArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M2.7Terminal-Bench 2.1 — Vals AI Terminus 248.7%vals.aiArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardVals AIIndependentRecord140002
MiniMax M3Terminal-Bench 2.1 — Vals AI Terminus 253.6%vals.aiArchived copyTerminal-Bench AccuracyReported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M2.5SWE-bench Verified — Vals AI mini-swe-agent74.2%vals.aiArchived copyResolved TasksReported Benchmark leaderboardVals AIIndependentRecord140001
MiniMax M2.5GPQA Diamond — Vals AI harness82.1%vals.aiArchived copyGPQA Diamond AccuracyReported Benchmark leaderboardVals AIIndependentRecord140001
MiniMax M2.5MMLU-Pro Original — Vals AI harness80.1%vals.aiArchived copyAccuracyReported Benchmark leaderboardVals AIIndependentRecord140001
MiniMax M2.7GPQA Diamond — Vals AI harness86.6%vals.aiArchived copyGPQA Diamond AccuracyReported Benchmark leaderboardVals AIIndependentRecord140002
MiniMax M2.7SWE-bench Verified — Vals AI mini-swe-agent73.8%vals.aiArchived copyResolved TasksReported Benchmark leaderboardVals AIIndependentRecord140002
MiniMax M2.7MMLU-Pro Original — Vals AI harness80.4%vals.aiArchived copyAccuracyReported Benchmark leaderboardVals AIIndependentRecord140002
MiniMax M3SWE-bench Verified — Vals AI mini-swe-agent75.0%vals.aiArchived copyResolved TasksReported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M3MMMU-Pro Vals AI harness81.2%vals.aiArchived copyMMMU-Pro accuracySetting: Standard (4 options)Reported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M3MMLU-Pro Original — Vals AI harness84.2%vals.aiArchived copyAccuracyReported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M3GPQA Diamond — Vals AI harness92.7%vals.aiArchived copyGPQA Diamond AccuracyReported Benchmark leaderboardVals AIIndependentRecord140003
MiniMax M2.7Toolathlon Verified47.5%toolathlon.xyzArchived copyTasks completedReported Benchmark leaderboardToolathlonIndependentToolathlon AuthorsRecord140002

MiniMax models