DeepSeek V4 vs MiMo V2.5: Benchmark Comparison
DeepSeek V4 has the higher score on 14 of 16 shared benchmarks; MiMo V2.5 leads on 1.
The largest observed score gap is 29.34 pts on Tax Agent Bench , where DeepSeek V4 leads.
Reported ±1 standard-error ranges overlap on 5 of 16 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 | DeepSeek V4 | MiMo V2.5 | Gap | Reported uncertainty |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | 3.75% ±1.43 | 1.67% ±0.00 | 2.08 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 23.08% ±2.93 | 9.13% ±2.00 | 13.94 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 80.32% ±0.47 | 78.89% ±0.51 | 1.44 pts | Reported ±1 SE ranges do not overlap |
| EMB | 51.62% ±3.05 | 55.09% ±3.13 | 3.47 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 44.08% ±0.65 | 36.73% ±0.11 | 7.35 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 58.50% ±3.13 | 29.16% ±2.71 | 29.34 pts | Reported ±1 SE ranges do not overlap |
| TaxEval v2 | 72.08% ±0.88 | 71.83% ±0.88 | 0.25 pts | Reported ±1 SE ranges overlap |
| MedCode | 40.45% ±2.12 | 31.89% ±2.02 | 8.56 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 75.14% ±2.00 | 72.15% ±1.85 | 2.99 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 16.00% ±3.69 | 16.00% ±3.69 | 0.00 pts | Reported ±1 SE ranges overlap |
| GPQA Diamond | 89.39% ±1.65 | 81.57% ±2.05 | 7.83 pts | Reported ±1 SE ranges do not overlap |
| MMLU Pro | 87.25% ±0.34 | 82.93% ±0.37 | 4.32 pts | Reported ±1 SE ranges do not overlap |
| Code Migration | 26.20% ±4.04 | 14.23% ±3.69 | 11.97 pts | Reported ±1 SE ranges do not overlap |
| LiveCodeBench | 87.48% ±0.95 | 81.51% ±1.07 | 5.97 pts | Reported ±1 SE ranges do not overlap |
| SWE-bench | 77.40% ±1.87 | 71.00% ±2.03 | 6.40 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 49.93% ±4.77 | 42.17% ±4.57 | 7.76 pts | Reported ±1 SE ranges overlap |
Performance by category
| Category | DeepSeek V4 average | MiMo V2.5 average |
|---|---|---|
| Legal | 35.72% | 29.90% |
| Finance | 56.57% | 48.20% |
| Healthcare | 57.80% | 52.02% |
| Math | 16.00% | 16.00% |
| Academic | 88.32% | 82.25% |
| Coding | 60.25% | 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 | DeepSeek V4 cost | MiMo V2.5 cost | DeepSeek V4 latency | MiMo V2.5 latency |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | $0.74 | $0.04 | 6m42s | 6m59s |
| Legal Research Bench | $0.68 | $0.03 | 17m05s | 6m31s |
| LegalBench | N/A | N/A | 3m04s | 11.18s |
| EMB | $0.80 | $0.08 | 13m42s | 30m38s |
| Finance Agent (v2) | $0.88 | $0.09 | 25m26s | 5m39s |
| Tax Agent Bench | $0.72 | $0.02 | 20m45s | 8m21s |
| TaxEval v2 | N/A | N/A | 3m10s | 26.30s |
| MedCode | N/A | N/A | 6m23s | 17.70s |
| MedScribe | N/A | N/A | 5m46s | 20.26s |
| ProofBench v1.1 | $0.03 | $0.06 | 7m03s | 20m56s |
| GPQA Diamond | N/A | N/A | 15m00s | 89.39s |
| MMLU Pro | N/A | N/A | 9m38s | 30.25s |
| Code Migration | $1.07 | $0.10 | 28m58s | 37m55s |
| LiveCodeBench | N/A | N/A | 11m56s | 108.61s |
| SWE-bench | $0.44 | $0.01 | 10m35s | 4m04s |
| Vibe Code Bench v1.1 | $2.21 | $0.07 | 56m59s | 26m20s |
Results available only for DeepSeek V4
- Vals Index
- ProgramBench
- SkillsBench
- Terminal-Bench 4.0
- Public Benefits Bench v1.1
Results available only for MiMo V2.5
- MortgageTax
- MMMU Pro
- SAGE