- SWE-Bench Verified 80.2
- SWE-Bench Pro 55.4
- Terminal Bench 2 51.7
- Multi-SWE-Bench 51.3
- SWE-Bench Multilingual 74.1
- VIBE-Pro (AVG) 54.2
- BrowseComp (w/ctx) 76.3
- Wide Search 70.3
- RISE 50.2
- BFCL multi-turn 76.8
- τ² Telecom 97.8
- GDPval-MM 59
- MEWC 74.4
- Finance Modeling 21.6
- AIME25 86.3
- GPQA-D 85.2
- HLE w/o tools 19.4
- SciCode 44.4
- IFBench 70
- AA-LCR 69.5
MiniMax-M2.5
MiniMax-M2.5 places in the top three on 2 benchmarks.
74
Benchmarks with a score
0
Ranked first, out of five or more models
2
In the top three
MiniMax Feb 2026 MiniMax-M2.5
Open weights Flagship
146.7
#79 of 271
Epoch Capabilities Index
- IFBench 72% #34 /70
- Terminal-Bench 2.0 42.7% #23 /43
- SciCode 44.4% #21 /47
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 | |
|---|---|---|---|---|
| MMLU-Pro | 85.2% | #24 of 151 | Model card | |
| AA-LCR (Artificial Analysis Long Context Reasoning) | 69.5% | #8 of 31 | Model card | |
| LMArena (Elo) | 1,391 | #140 of 333 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 9.10% | #46 of 104 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 65.3 | #54 of 119 | Official leaderboard | |
| IFBench | 72% | #34 of 70 | Model card | |
| LOCA-Bench | 6.70% | #5 of 9 | Official leaderboard | |
| WideSearch | 70.3% | #11 of 17 | Model card | |
| Judgemark v2.1 | 64.5 | #28 of 42 | Official leaderboard | |
| Humanity's Last Exam | 19.4% | #100 of 147 | Model card | |
| Creative Writing v3 | 60.2 | #82 of 120 | Official leaderboard | |
| BullshitBench | 16.4% | #78 of 107 | Official leaderboard | |
| Extended NYT Connections | 16.8% | #84 of 97 | Official leaderboard | |
| BullshitBench V2 | 9.00% | #109 of 122 | Official leaderboard | |
| CL-bench | 11.4% | #24 of 24 | Official leaderboard | |
| SkateBench v2 | 13.8% | #21 of 21 | Official leaderboard | |
| CL-bench Life | 6.30% | #16 of 16 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 85.2% | #72 of 284 | Model card | |
| AIME 2025 | 86.3% | #67 of 143 | Model card | |
| ARC-AGI | 63.7% | #56 of 92 | Official leaderboard | |
| ARC-AGI-2 | 4.86% | #68 of 95 | Official leaderboard | |
| Echo General AI Prediction Leaderboard | 980 | #14 of 17 | Official leaderboard | |
| AIME 2026 | 87.2% | #31 of 34 | Model card | |
| ProofBench | 4.00% | #54 of 59 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 75.8% | #5 of 60 | Official leaderboard | |
| Multi-SWE-bench | 51.3% | #2 of 10 | Model card | |
| MLE-bench | 51.5% | #4 of 18 | Model card | |
| SWE-bench Multilingual | 74.1% | #30 of 70 | Model card | |
| SciCode | 44.4% | #21 of 47 | Model card | |
| SWE-Bench Pro | 55.4% | #46 of 77 | Model card | |
| OpenHands Index | 45.2% | #22 of 33 | Official leaderboard | |
| VIBE-Pro (MiniMax) | 54.2% | #2 of 3 | Model card | |
| IOI (Vals) | 6.67% | #37 of 55 | Official leaderboard | |
| WebDev Arena | 1,384 | #73 of 105 | Epoch AI | |
| Code Arena | 1,386 | #88 of 124 | Official leaderboard | |
| ALE-Bench | 618 | #92 of 121 | Epoch AI | |
| Vibe Code Bench | 14.8% | #90 of 115 | Official leaderboard | |
| PostTrainBench | 9.50% | #19 of 23 | Official leaderboard | |
| FrontierCode | 5.34% | #11 of 12 | Official leaderboard | |
| SWE Atlas Refactoring | 19.5% | #16 of 17 | Official leaderboard | |
| SWE Atlas | 10.3% | #21 of 22 | Official leaderboard | |
| NL2Repo-Bench | 26.6% | #28 of 28 | Model card | |
| SWE Atlas Test Writing | 18.6% | #22 of 22 | Official leaderboard | |
| OJBench | 41.4% | #8 of 8 | Official leaderboard | |
| HyperTask | 59.4% | #2 of 2 | Model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| tau2-bench | 97.8% | #3 of 69 | Model card | |
| Berkeley Function Calling Leaderboard (BFCL) | 76.8% | #7 of 47 | Model card | |
| SnakeBench | 27 | #54 of 183 | Official leaderboard | |
| LisanBench | 852.5 | #58 of 133 | Official leaderboard | |
| Toolathlon | 38.3% | #22 of 50 | Official leaderboard | |
| BrowseComp | 76.3% | #37 of 77 | Model card | |
| Pencil Puzzle Bench | 3.30% | #28 of 53 | Official leaderboard | |
| Terminal-Bench 2.0 | 42.7% | #23 of 43 | Official leaderboard | |
| PinchBench | 67.2% | #26 of 48 | Official leaderboard | |
| Claw-Eval-Live | 50.5% | #10 of 12 | Official leaderboard | |
| ClawBench V2 | 1.50% | #7 of 7 | Official leaderboard | |
| MM-ClawBench | 57.6% | #2 of 2 | Model card | |
| RISE | 50.2% | #2 of 2 | Model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CorpFin v2 | 59.6% | #74 of 119 | Official leaderboard | |
| LegalBench | 80% | #87 of 138 | Official leaderboard | |
| TaxEval v2 | 68.2% | #94 of 129 | Official leaderboard | |
| CaseLaw v2 | 53.5% | #40 of 53 | Official leaderboard | |
| ProfBench | 42.2% | #57 of 72 | Official leaderboard | |
| Vals Finance Agent | 38.6% | #40 of 49 | Official leaderboard | |
| Vending-Bench 2 | $-23 | #65 of 66 | Official leaderboard | |
| SpreadsheetBench 2 | 7.17% | #8 of 8 | Official leaderboard | |
| Finance Modeling Pro | 33.8% | #2 of 2 | Model card | |
| MEWC v2 | 49.8% | #2 of 2 | Model card | |
| Finance Modeling | 21.6% | #1 of 1 | Model card | |
| GDPval-MM | 59% | #1 of 1 | Model card | |
| MEWC | 74.4% | #1 of 1 | Model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SEC-bench Pro | 0% | #6 of 6 | Official leaderboard | |
| SEC-bench Pro 260617 | 0.60% | #6 of 6 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Artificial Analysis Intelligence Index v4.0 | 42.0 | #4 of 5 | Model card |
MiniMax's models over time
MiniMax's models in color, against every other model on the Epoch Capabilities Index, by release date.
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.
Its system card
Every result in the longer documents the lab published with it. 11 of them aren't in the launch post.
Capabilities
- SWE-Bench Pro 55.4
- SWE-bench Multilingual 74.1
- Multi-SWE-bench 51.3
- NL2Repo-Bench 26.6
- Terminal-Bench 2.0 51.7
- VIBE-Pro (MiniMax) 54.2
- HyperTask 59.4
- BrowseComp 76.3
- WideSearch 70.3
- RISE 50.2
- GDPval-AA 35
- Toolathlon 38.3
- MM-ClawBench 57.6
- MEWC v2 49.8
- Finance Modeling Pro 33.8
- AIME 2026 87.2
- GPQA Diamond 85.2
- SciCode 43
- IFBench 72
- AA-LCR (Artificial Analysis Long Context Reasoning) 65
- Humanity's Last Exam 19
- MMLU-Pro 85.2
- AIME 2025 86.3
- Artificial Analysis Intelligence Index v4.0 42
AI R&D risk
- MLE-bench 51.5