Kimi K2.6 vs MiniMax-M3: Benchmark Comparison
Kimi K2.6 has the higher score on 7 of 19 shared benchmarks; MiniMax-M3 leads on 12.
The largest observed score gap is 13.94 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 | Kimi K2.6 | MiniMax-M3 | Gap | Reported uncertainty |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | 1.67% ±0.83 | 4.17% ±1.65 | 2.50 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 15.87% ±2.54 | 29.81% ±3.18 | 13.94 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 84.74% ±0.45 | 85.42% ±0.44 | 0.68 pts | Reported ±1 SE ranges overlap |
| EMB | 57.85% ±2.74 | 47.76% ±2.95 | 10.10 pts | Reported ±1 SE ranges do not overlap |
| Finance Agent (v2) | 44.90% ±0.73 | 48.27% ±0.44 | 3.37 pts | Reported ±1 SE ranges do not overlap |
| MortgageTax | 65.82% ±0.93 | 68.36% ±0.91 | 2.54 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 45.13% ±3.23 | 49.69% ±3.25 | 4.56 pts | Reported ±1 SE ranges overlap |
| TaxEval v2 | 74.65% ±0.85 | 72.73% ±0.86 | 1.92 pts | Reported ±1 SE ranges do not overlap |
| MedCode | 40.14% ±2.04 | 46.29% ±2.10 | 6.15 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 78.15% ±1.79 | 87.25% ±1.96 | 9.10 pts | Reported ±1 SE ranges do not overlap |
| GPQA Diamond | 89.14% ±2.00 | 92.68% ±1.44 | 3.53 pts | Reported ±1 SE ranges do not overlap |
| MMLU Pro | 87.57% ±0.33 | 84.22% ±0.36 | 3.35 pts | Reported ±1 SE ranges do not overlap |
| MMMU Pro | 86.30% ±0.83 | 81.16% ±0.94 | 5.14 pts | Reported ±1 SE ranges do not overlap |
| SAGE | 50.22% ±3.43 | 50.57% ±3.44 | 0.35 pts | Reported ±1 SE ranges overlap |
| Code Migration | 27.77% ±4.14 | 19.93% ±3.94 | 7.84 pts | Reported ±1 SE ranges overlap |
| LiveCodeBench | 86.77% ±0.97 | 82.15% ±1.05 | 4.62 pts | Reported ±1 SE ranges do not overlap |
| SWE-bench | 76.20% ±1.91 | 75.00% ±1.94 | 1.20 pts | Reported ±1 SE ranges overlap |
| Vibe Code Bench v1.1 | 37.89% ±4.91 | 47.57% ±5.44 | 9.68 pts | Reported ±1 SE ranges overlap |
| Public Benefits Bench v1.1 | 56.63% ±1.29 | 64.14% ±1.25 | 7.51 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Kimi K2.6 average | MiniMax-M3 average |
|---|---|---|
| Legal | 34.09% | 39.80% |
| Finance | 57.67% | 57.36% |
| Healthcare | 59.15% | 66.77% |
| Academic | 87.67% | 86.02% |
| Education | 50.22% | 50.57% |
| Coding | 57.16% | 56.16% |
| Social Mobility | 56.63% | 64.14% |
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 | Kimi K2.6 cost | MiniMax-M3 cost | Kimi K2.6 latency | MiniMax-M3 latency |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | $0.82 | $1.46 | 23m23s | 22m31s |
| Legal Research Bench | $1.52 | $0.34 | 30m53s | 13m34s |
| LegalBench | N/A | N/A | 50.07s | 8.21s |
| EMB | $2.20 | $2.09 | 39m48s | 31m30s |
| Finance Agent (v2) | $1.14 | $0.32 | 27m10s | 8m17s |
| MortgageTax | N/A | N/A | 109.76s | 26.76s |
| Tax Agent Bench | $0.57 | $0.16 | 15m24s | 5m15s |
| TaxEval v2 | N/A | N/A | 4m27s | 94.04s |
| MedCode | N/A | N/A | 5m05s | 63.12s |
| MedScribe | N/A | N/A | 8m14s | 2m04s |
| GPQA Diamond | N/A | N/A | 9m05s | 4m39s |
| MMLU Pro | N/A | N/A | 2m36s | 41.39s |
| MMMU Pro | N/A | N/A | 4m02s | 68.35s |
| SAGE | N/A | N/A | 4m31s | 118.54s |
| Code Migration | $5.11 | $7.07 | 1h30m | 1h14m |
| LiveCodeBench | N/A | N/A | 8m49s | 6m07s |
| SWE-bench | $0.27 | $0.42 | 8m40s | 12m06s |
| Vibe Code Bench v1.1 | $1.93 | $6.45 | 49m28s | 1h25m |
| Public Benefits Bench v1.1 | $0.91 | $0.25 | 9m55s | 11m31s |
Results available only for Kimi K2.6
- ProgramBench
Results available only for MiniMax-M3
- Vals Index
- ProofBench v1.1
- SkillsBench
- Terminal-Bench 4.0
- Vibe Code Bench 1-100
- CyberBench v1.1