MiniMax-M3 vs MiMo V2.5: Benchmark Comparison
MiniMax-M3 has the higher score on 18 of 19 shared benchmarks; MiMo V2.5 leads on 1.
The largest observed score gap is 20.67 pts on Legal Research Bench , where MiniMax-M3 leads.
Reported ±1 standard-error ranges overlap on 6 of 19 shared benchmarks with comparable uncertainty data. This is not a pairwise statistical significance test.
Shared benchmark results
Scores and reported standard errors come from the latest visible Vals benchmark version available for both models.
| Benchmark | MiniMax-M3 | MiMo V2.5 | Gap | Reported uncertainty |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | 4.17% ±1.65 | 1.67% ±0.00 | 2.50 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 29.81% ±3.18 | 9.13% ±2.00 | 20.67 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 85.42% ±0.44 | 78.89% ±0.51 | 6.53 pts | Reported ±1 SE ranges do not overlap |
| EMB | 47.76% ±2.95 | 55.09% ±3.13 | 7.33 pts | Reported ±1 SE ranges do not overlap |
| Finance Agent (v2) | 48.27% ±0.44 | 36.73% ±0.11 | 11.54 pts | Reported ±1 SE ranges do not overlap |
| MortgageTax | 68.36% ±0.91 | 59.26% ±0.98 | 9.10 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 49.69% ±3.25 | 29.16% ±2.71 | 20.53 pts | Reported ±1 SE ranges do not overlap |
| TaxEval v2 | 72.73% ±0.86 | 71.83% ±0.88 | 0.90 pts | Reported ±1 SE ranges overlap |
| MedCode | 46.29% ±2.10 | 31.89% ±2.02 | 14.39 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 87.25% ±1.96 | 72.15% ±1.85 | 15.10 pts | Reported ±1 SE ranges do not overlap |
| ProofBench v1.1 | 18.00% ±3.86 | 16.00% ±3.69 | 2.00 pts | Reported ±1 SE ranges overlap |
| GPQA Diamond | 92.68% ±1.44 | 81.57% ±2.05 | 11.11 pts | Reported ±1 SE ranges do not overlap |
| MMLU Pro | 84.22% ±0.36 | 82.93% ±0.37 | 1.29 pts | Reported ±1 SE ranges do not overlap |
| MMMU Pro | 81.16% ±0.94 | 80.00% ±0.96 | 1.16 pts | Reported ±1 SE ranges overlap |
| SAGE | 50.57% ±3.44 | 43.27% ±3.39 | 7.31 pts | Reported ±1 SE ranges do not overlap |
| Code Migration | 19.93% ±3.94 | 14.23% ±3.69 | 5.70 pts | Reported ±1 SE ranges overlap |
| LiveCodeBench | 82.15% ±1.05 | 81.51% ±1.07 | 0.64 pts | Reported ±1 SE ranges overlap |
| SWE-bench | 75.00% ±1.94 | 71.00% ±2.03 | 4.00 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 47.57% ±5.44 | 42.17% ±4.57 | 5.39 pts | Reported ±1 SE ranges overlap |
Performance by category
| Category | MiniMax-M3 average | MiMo V2.5 average |
|---|---|---|
| Legal | 39.80% | 29.90% |
| Finance | 57.36% | 50.41% |
| Healthcare | 66.77% | 52.02% |
| Math | 18.00% | 16.00% |
| Academic | 86.02% | 81.50% |
| Education | 50.57% | 43.27% |
| Coding | 56.16% | 52.23% |
Cost and latency
Cost per test appears only where the benchmark reports it for both models. Latency is the measured completion time for that benchmark.
| Benchmark | MiniMax-M3 cost | MiMo V2.5 cost | MiniMax-M3 latency | MiMo V2.5 latency |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | $1.46 | $0.04 | 22m31s | 6m59s |
| Legal Research Bench | $0.34 | $0.03 | 13m34s | 6m31s |
| LegalBench | N/A | N/A | 8.21s | 11.18s |
| EMB | $2.09 | $0.08 | 31m30s | 30m38s |
| Finance Agent (v2) | $0.32 | $0.09 | 8m17s | 5m39s |
| MortgageTax | N/A | N/A | 26.76s | 32.94s |
| Tax Agent Bench | $0.16 | $0.02 | 5m15s | 8m21s |
| TaxEval v2 | N/A | N/A | 94.04s | 26.30s |
| MedCode | N/A | N/A | 63.12s | 17.70s |
| MedScribe | N/A | N/A | 2m04s | 20.26s |
| ProofBench v1.1 | $0.42 | $0.06 | 11m59s | 20m56s |
| GPQA Diamond | N/A | N/A | 4m39s | 89.39s |
| MMLU Pro | N/A | N/A | 41.39s | 30.25s |
| MMMU Pro | N/A | N/A | 68.35s | 42.47s |
| SAGE | N/A | N/A | 118.54s | 52.66s |
| Code Migration | $7.07 | $0.10 | 1h14m | 37m55s |
| LiveCodeBench | N/A | N/A | 6m07s | 108.61s |
| SWE-bench | $0.42 | $0.01 | 12m06s | 4m04s |
| Vibe Code Bench v1.1 | $6.45 | $0.07 | 1h25m | 26m20s |
Results available only for MiniMax-M3
- Vals Index
- SkillsBench
- Terminal-Bench 4.0
- Vibe Code Bench 1-100
- CyberBench v1.1
- Public Benefits Bench v1.1
Results available only for MiMo V2.5
None.