- SWE-Bench Verified 80.5
- SWE-Bench Pro 59
- Terminal Bench 2.1 66
- SWE Atlas-QnA 37.9
- nl2repo 42.1
- SWE Atlas-Test Writing 30.8
- SWE-fficiency 34.8
- LiveSQLBench 40.2
- CL-bench 20.5
- VIBE-V2 50.1
- SVG-Bench 63.7
- PostTrainBench 37.1
- KernelBench Hard 28.8
- PaperBench 52.6
- BrowseComp 83.5
- DRACO 73.2
- GDPval rubrics 74.8
- BankerToolBench 76.1
- OfficeQA Pro 45.1
- SpreadSheetBench-v1 89.3
- YC-Bench 2100000
- LOCA-Bench (256k) 49.3
- MCP Atlas 74.2
- Apex-Agents 27.7
- Claw-Eval 74.5
- OSWorld-Verified 75.2
- OmniDocBench 91.6
- MMMU-Pro 78.1
- Video-MMMU 84.6
- VideoMME (w/ sub) 85.4
- IMO 2025 35
- USAMO 2026 36
MiniMax-M3
MiniMax-M3 ranks first on 2 of the 182 benchmarks with five or more models scored.
191
Benchmarks with a score
2
Ranked first, out of five or more models
6
In the top three
MiniMax Jun 2026 MiniMax-M3
Open weights Flagship
147.7
#79 of 256
Epoch Capabilities Index
Self-Reported Index 151.3 #85 of 404
- MMMU-Pro 78.1% #27 /90
- AutomationBench 4.87% #33 /42
- NL2Repo-Bench 42.1% #22 /30
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| StudyArena Text | 1,347 | #2 of 85 | Official leaderboard | |
| BullshitBench V2 | 64% | #31 of 129 | Official leaderboard | |
| LMArena (Elo) | 1,436 | #82 of 335 | Official leaderboard | |
| BullshitBench | 47.3% | #30 of 107 | Official leaderboard | |
| SEA-HELM | 72.3 | #26 of 71 | Official leaderboard | |
| Wer wird Millionär? LLM Benchmark | 14756.0 | #31 of 77 | Official leaderboard | |
| Last Translation Benchmark | 6.26 | #19 of 36 | Official leaderboard | |
| Lech Mazur Creative Writing | -0.14 | #28 of 50 | Official leaderboard | |
| DesignArena ASCII Art | 1,149 | #46 of 77 | Official leaderboard | |
| Extended NYT Connections | 65.1% | #63 of 103 | Official leaderboard | |
| CL-bench | 20.5% | #4 of 6 | Lab model card | |
| EQ-Bench 4 | 1,140 | #24 of 32 | Official leaderboard | |
| AA-LCR Extended | 45.5% | #38 of 49 | Official leaderboard | |
| CompassArena | 990 | #8 of 10 | Official leaderboard | |
| Document Arena | 1,435 | #34 of 39 | Official leaderboard | |
| OpenAI MRCR v2 | 18.6% | #41 of 47 | Official leaderboard | |
| HLE Extended | 5.40% | #29 of 31 | Official leaderboard | |
| LOCA-Bench | 49.3% | #2 of 2 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 90.9% | #39 of 229 | Epoch AI run | |
| ResearchClawBench | 19.8 | #9 of 23 | Official leaderboard | |
| ForecastBench | 0.13 | #36 of 82 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 71.1% | #107 of 212 | Epoch AI run | |
| Surface Evolver Bench | 55% | #13 of 25 | Official leaderboard | |
| DTBench | 78.9% | #96 of 180 | Official leaderboard | |
| ProofBench | 19% | #32 of 59 | Official leaderboard | |
| SimpleBench | 45.8% | #67 of 106 | Official leaderboard | |
| ProofBench v1.1 | 18% | #42 of 52 | Official leaderboard | |
| ASI-Bench | 26.6 | #9 of 11 | Official leaderboard | |
| NatureBench | 6.67% | #16 of 18 | Official leaderboard | |
| Science's Last Exam | 16.8 | #9 of 9 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveSQLBench | 40.2% | #1 of 36 | Official leaderboard | |
| SWE-bench Verified | 80.5% | #14 of 155 | Lab model card | |
| EvoCode-Bench v2 | 23.4% | #3 of 13 | Official leaderboard | |
| PaperBench | 52.6% | #4 of 15 | Lab model card | |
| Monthly-SWEBench | 58% | #4 of 12 | Official leaderboard | |
| SWE-Bench Pro | 59% | #30 of 81 | Lab model card | |
| DesignArena Website | 1,249 | #56 of 151 | Official leaderboard | |
| OpenHands Index | 57.2% | #13 of 33 | Official leaderboard | |
| DesignArena Code | 1,237 | #57 of 143 | Official leaderboard | |
| DesignArena UI Component | 1,236 | #57 of 139 | Official leaderboard | |
| DesignArena Data Visualization | 1,231 | #50 of 118 | Official leaderboard | |
| WebDev Arena | 1,487 | #46 of 105 | arena.ai, via Epoch AI | |
| DesignArena Agentic Overall | 1,180 | #33 of 74 | Official leaderboard | |
| DesignArena Game Dev | 1,201 | #64 of 143 | Official leaderboard | |
| DesignArena Full Stack | 1,179 | #22 of 49 | Official leaderboard | |
| DesignArena 3D Design | 1,192 | #63 of 136 | Official leaderboard | |
| DesignArena Frontend Web App | 1,192 | #26 of 53 | Official leaderboard | |
| Code Arena | 1,484 | #63 of 128 | Official leaderboard | |
| SWE-rebench | 47.2% | #7 of 13 | Official leaderboard | |
| DesignArena SVG | 1,147 | #57 of 103 | Official leaderboard | |
| DesignArena React Native | 1,171 | #31 of 56 | Official leaderboard | |
| PostTrainBench | 37.1% | #4 of 7 | Lab model card | |
| Vision2Web S3-2026 | 59% | #4 of 7 | Official leaderboard | |
| Vibe Code Bench | 47.6% | #69 of 120 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,156 | #28 of 46 | Official leaderboard | |
| AndroidBench | 63.6% | #20 of 32 | Official leaderboard | |
| Claw-SWE-Bench | 69.7% | #10 of 16 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 10% | #24 of 38 | Official leaderboard | |
| DesignArena Image to Frontend | 1,156 | #20 of 31 | Official leaderboard | |
| APEX-SWE | 34.6% | #44 of 65 | Official leaderboard | |
| Code Migration | 19.9% | #53 of 78 | Official leaderboard | |
| Senior SWE-Bench | 21.1% | #16 of 23 | Official leaderboard | |
| DesignArena Agentic Game Dev | 1,115 | #19 of 27 | Official leaderboard | |
| Image-to-WebDev Arena | 1,473 | #40 of 55 | Official leaderboard | |
| NL2Repo-Bench | 42.1% | #22 of 30 | Lab model card | |
| EvoPolicyGym | 0.53 | #3 of 4 | Official leaderboard | |
| GBAEval | 0.91% | #16 of 21 | Official leaderboard | |
| Next.js Agent Evals | 82% | #11 of 14 | Official leaderboard | |
| DesignArena Video to Website | 1,163 | #15 of 19 | Official leaderboard | |
| ALE-Bench | 1,053 | #100 of 126 | Official leaderboard | |
| SWE-bench Verified Extended | 28.7% | #35 of 44 | Official leaderboard | |
| DAXBench v2 | 64.2% | #38 of 47 | Official leaderboard | |
| DesignArena Image to HTML | 1,151 | #43 of 51 | Official leaderboard | |
| SWE Atlas | 37.9% | #12 of 14 | Lab model card | |
| SWE Atlas v1.1 | 39.3% | #25 of 28 | Official leaderboard | |
| Code Migration v1.1 | 20.9% | #27 of 30 | Official leaderboard | |
| DesignArena Blender | 1,090 | #11 of 12 | Official leaderboard | |
| Vibe Code Bench 1-100 | 9.22% | #24 of 26 | Official leaderboard | |
| Tencent WorkBuddy Bench | 65.5% | #14 of 14 | Official leaderboard | |
| KernelBench | 28.8% | #1 of 1 | Lab model card | |
| SVG-Bench (MiniMax) | 63.7% | #1 of 1 | Lab model card | |
| SWE-fficiency | 34.8% | #1 of 1 | Lab model card | |
| VIBE-V2 (MiniMax) | 50.1% | #1 of 1 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ClawProBench | 73.2% | #3 of 49 | Official leaderboard | |
| OSWorld-Verified | 75.2% | #9 of 33 | Official leaderboard | |
| BrowseComp | 83.5% | #23 of 81 | Lab model card | |
| LiveClawBench | 71.4% | #5 of 17 | Official leaderboard | |
| Claw-Eval | 74.5% | #4 of 13 | Lab model card | |
| Pencil Puzzle Bench | 7.10% | #18 of 53 | Official leaderboard | |
| SkillsBench | 53% | #9 of 22 | Official leaderboard | |
| Long-Horizon-Terminal-Bench (LHTB) | 38.5% | #11 of 25 | Official leaderboard | |
| SnakeBench | 25 | #87 of 188 | Official leaderboard | |
| TUA-Bench | 47% | #6 of 12 | Official leaderboard | |
| MCP Atlas | 74.2% | #25 of 46 | Lab model card | |
| GameHorizon | 65.6 | #21 of 37 | Official leaderboard | |
| Long-Horizon-Terminal-Bench 2.0 | 46.9% | #10 of 15 | Official leaderboard | |
| Harbor-Index | 6.10% | #6 of 9 | Official leaderboard | |
| Terminal-Bench 2.1 | 66% | #40 of 56 | Lab model card | |
| AutomationBench | 4.87% | #33 of 42 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 22.2% | #35 of 44 | Official leaderboard | |
| Chess Puzzles | 17% | #76 of 93 | Epoch AI run | |
| CivBench | 1,448 | #16 of 19 | Official leaderboard | |
| BabaIsBench | 46% | #27 of 32 | Official leaderboard | |
| Terminal-Bench 4.0 | 7.27% | #34 of 40 | Official leaderboard | |
| OSWorld 2.0 | 4.60% | #6 of 7 | Official leaderboard | |
| Terminal-Bench 4.1 | 7.58% | #37 of 43 | Official leaderboard | |
| Mystery Game Puzzles | 8.00% | #74 of 84 | Epoch AI run | |
| Agent Arena | -10.46% | #50 of 56 | Official leaderboard | |
| BrowseComp Extended | 7.70% | #38 of 42 | Official leaderboard | |
| DRACO | 73.2% | #12 of 12 | Lab model card | |
| OSWorld-Science | 13% | #11 of 11 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CorpFin v2 | 68.1% | #11 of 119 | Official leaderboard | |
| CorpBench Work | 89.6 | #2 of 17 | Official leaderboard | |
| LegalBench | 85.4% | #21 of 138 | Official leaderboard | |
| SpreadsheetBench | 89.3% | #1 of 6 | Lab model card | |
| MortgageTax | 68.4% | #15 of 87 | Official leaderboard | |
| MedScribe | 87.3% | #18 of 97 | Official leaderboard | |
| DesignArena PDF | 1,260 | #7 of 37 | Official leaderboard | |
| DesignArena Spreadsheet | 1,246 | #11 of 35 | Official leaderboard | |
| MedCode | 46.3% | #36 of 96 | Official leaderboard | |
| FORTE | 71.2% | #8 of 21 | Official leaderboard | |
| TaxEval v2 | 72.7% | #58 of 129 | Official leaderboard | |
| LEXam-hard | 37.1% | #7 of 15 | Official leaderboard | |
| DesignArena Python-PPTX Slides | 1,171 | #17 of 36 | Official leaderboard | |
| Public Benefits Bench | 64.1% | #24 of 50 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 4.17% | #39 of 80 | Official leaderboard | |
| DesignArena HTML Slides | 1,077 | #16 of 28 | Official leaderboard | |
| YC-Bench | $999,506 | #36 of 59 | Official leaderboard | |
| Legal Research Bench | 29.8% | #58 of 93 | Official leaderboard | |
| Vals Finance Agent v2 | 48.3% | #51 of 80 | Official leaderboard | |
| Vals Index v2 | 42.7% | #42 of 65 | Official leaderboard | |
| Tax Agent Bench | 20.6% | #56 of 84 | Official leaderboard | |
| ProfBench | 44.4% | #48 of 72 | Official leaderboard | |
| RadLE 2.0 | 350.0 | #8 of 12 | Official leaderboard | |
| EnterpriseClawBench | 52.5% | #6 of 9 | Official leaderboard | |
| Genspark Slides Benchmark | 56.1% | #7 of 10 | Official leaderboard | |
| GDPval Extended | 0.40% | #31 of 43 | Official leaderboard | |
| Vending-Bench 2 | $2,158 | #50 of 67 | Official leaderboard | |
| APEX-1 | 59.8% | #53 of 71 | Official leaderboard | |
| Excel Modeling Benchmark | 47.8% | #57 of 76 | Official leaderboard | |
| APEX-Agents | 27.7% | #9 of 12 | Lab model card | |
| APEX-Agents 1.1 | 53.5% | #53 of 65 | Official leaderboard | |
| FoodTruck Bench | $882 | #49 of 59 | Official leaderboard | |
| Vals Finance Index | 48.4% | #35 of 40 | Official leaderboard | |
| Vals Index v2.1 | 36.5% | #43 of 49 | Official leaderboard | |
| PolicyBench | 75.4% | #44 of 50 | Official leaderboard | |
| APEX-Accounting | 41.8% | #44 of 49 | Official leaderboard | |
| Vals Finance Agent v2.1 | 54.3% | #36 of 40 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 59.8% | #38 of 42 | Official leaderboard | |
| Clipping Bench | $223 | #18 of 19 | Official leaderboard | |
| Immigration Bench | 57.5% | #24 of 25 | Official leaderboard | |
| MedXpertQA MM Extended | 34.6% | #41 of 41 | Official leaderboard | |
| Thrive Holdings helpdesk ticket resolution benchmark | 23.1% | #15 of 15 | Official leaderboard | |
| OfficeQA | 45.1% | #14 of 14 | Lab model card | |
| BlenderWorld | 31.8 | #13 of 13 | Official leaderboard | |
| BankerToolBench | 76.1% | #3 of 3 | Lab model card | |
| GDPval-Rubrics (MiniMax) | 74.8% | #1 of 1 | Lab model card | |
| SpreadsheetBench Verified | 82.5% | #1 of 1 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DesignArena Code to Video | 1,216 | #7 of 29 | Official leaderboard | |
| MMMU-Pro | 78.1% | #27 of 90 | Lab model card | |
| Video-MMMU | 84.6% | #11 of 36 | Lab model card | |
| Video-MME | 85.4% | #18 of 51 | Lab model card | |
| Vision Arena | 1,232 | #59 of 136 | Official leaderboard | |
| AgenticVBench | 22.7% | #5 of 9 | Official leaderboard | |
| VoxelBench | 1,408 | #45 of 56 | Official leaderboard | |
| BVB | 58.2 | #39 of 45 | Official leaderboard | |
| 4DCodeBench | 0.34 | #14 of 16 | Official leaderboard | |
| PerceptionBench (Moonshot) | 33.1% | #15 of 16 | Official leaderboard | |
| CharXiv Extended | 58.4% | #38 of 38 | Official leaderboard | |
| MMMU-Pro Extended | 18.1% | #38 of 38 | Official leaderboard | |
| Code2Video-Compass | 15.0 | #15 of 15 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CyberBench (Vals) | 76.3% | #8 of 26 | Official leaderboard | |
| CyberGym | 73.1% | #15 of 34 | Official leaderboard | |
| nocot-bench (NCRI) | 72.6 | #85 of 185 | Official leaderboard | |
| SAFE-Teen Relationship boundaries | 53% | #9 of 17 | Official leaderboard | |
| SAFE-Teen Health advice | 49% | #10 of 17 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 65% | #30 of 44 | Official leaderboard | |
| DistressBench | 67.7% | #32 of 42 | Official leaderboard | |
| PatchEval | 66.1% | #6 of 7 | Official leaderboard | |
| PatchEval-Verified | 66.1% | #6 of 7 | Official leaderboard | |
| SAFE-Teen Urgent safety | 54% | #16 of 17 | Official leaderboard | |
| DrugDiscoveryBench | 22.8% | #20 of 20 | Official leaderboard | |
| VIALS 2 | 10.8% | #14 of 14 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SAGE (Vals) | 50.6% | #16 of 81 | Official leaderboard | |
| ACCoRD | 65.9 | #70 of 152 | Official leaderboard | |
| Conceptual Reasoning Index | 46.4 | #82 of 151 | Official leaderboard | |
| Vals Multimodal Index | 60% | #18 of 33 | Official leaderboard | |
| SAGE v1.1 (Vals) | 85.5% | #16 of 29 | Official leaderboard | |
| LMCA | 33.7% | #83 of 147 | Official leaderboard | |
| SuperCLUE | 41.2 | #12 of 16 | Official leaderboard | |
| CompassRank | 69.0 | #14 of 14 | Official leaderboard |
Its lab's numbers put it at 151.3, 3.6 points above its ECI
The Self-Reported Index fits one score to the 16 capability results MiniMax reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 149.1 to 153.6. How we compute it.
- Video-MME 85.4
- Video-MMMU 84.6
- BrowseComp 83.5
- SWE-bench Verified 80.5
- MMMU-Pro 78.1
- OSWorld-Verified 75.2
- Claw-Eval 74.5
- MCP Atlas 74.2
- DRACO 73.2
- Terminal-Bench 2.1 66
- SWE-Bench Pro 59
- PaperBench 52.6
- OfficeQA 45.1
- NL2Repo-Bench 42.1
- SWE Atlas 37.9
- APEX-Agents 27.7
Its lab's numbers next to 3 leaderboards
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| YC-Bench | $2,100,000 YC-Bench | $999,506 Leaderboard | +$1,100,494 |
| OSWorld-Verified | 75.2% OSWorld-Verified | 75.2% Leaderboard, 100 steps | Same |
| LiveSQLBench | 40.2% LiveSQLBench | 40.2% Leaderboard, Claude Code | Same |
MiniMax's models over time
MiniMax's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.