DeepSeek V4 vs Gemini 3.1 Pro Preview (02/26): Benchmark Comparison
DeepSeek V4 has the higher score on 9 of 20 shared benchmarks; Gemini 3.1 Pro Preview (02/26) leads on 10.
The largest observed score gap is 20.66 pts on Tax Agent Bench , where DeepSeek V4 leads.
Reported ±1 standard-error ranges overlap on 8 of 20 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 | Gemini 3.1 Pro Preview (02/26) | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Index | 38.63% ±1.16 | 33.44% ±1.13 | 5.19 pts | Reported ±1 SE ranges do not overlap |
| Harvey's Legal Agent Benchmark | 3.75% ±1.43 | 0.00% ±0.00 | 3.75 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 23.08% ±2.93 | 20.67% ±2.81 | 2.40 pts | Reported ±1 SE ranges overlap |
| LegalBench | 80.32% ±0.47 | 87.40% ±0.33 | 7.08 pts | Reported ±1 SE ranges do not overlap |
| EMB | 51.62% ±3.05 | 52.62% ±2.97 | 0.99 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 44.08% ±0.65 | 42.98% ±1.21 | 1.10 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 58.50% ±3.13 | 37.84% ±3.10 | 20.66 pts | Reported ±1 SE ranges do not overlap |
| TaxEval v2 | 72.08% ±0.88 | 72.88% ±0.86 | 0.81 pts | Reported ±1 SE ranges overlap |
| MedCode | 40.45% ±2.12 | 59.06% ±2.00 | 18.61 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 75.14% ±2.00 | 76.11% ±1.92 | 0.97 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 16.00% ±3.69 | 26.00% ±4.41 | 10.00 pts | Reported ±1 SE ranges do not overlap |
| GPQA Diamond | 89.39% ±1.65 | 95.45% ±1.05 | 6.06 pts | Reported ±1 SE ranges do not overlap |
| MMLU Pro | 87.25% ±0.34 | 90.99% ±0.28 | 3.74 pts | Reported ±1 SE ranges do not overlap |
| Code Migration | 26.20% ±4.04 | 17.31% ±3.93 | 8.89 pts | Reported ±1 SE ranges do not overlap |
| LiveCodeBench | 87.48% ±0.95 | 88.48% ±0.93 | 1.00 pts | Reported ±1 SE ranges overlap |
| ProgramBench | 0.00% ±0.00 | 0.00% ±0.00 | 0.00 pts | Reported ±1 SE ranges overlap |
| SWE-bench | 77.40% ±1.87 | 78.80% ±1.83 | 1.40 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 11.11% ±1.34 | 2.52% ±0.51 | 8.59 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 49.93% ±4.77 | 32.03% ±4.34 | 17.90 pts | Reported ±1 SE ranges do not overlap |
| Public Benefits Bench v1.1 | 62.92% ±1.26 | 53.79% ±1.30 | 9.13 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | DeepSeek V4 average | Gemini 3.1 Pro Preview (02/26) average |
|---|---|---|
| Legal | 35.72% | 36.02% |
| Finance | 56.57% | 51.58% |
| Healthcare | 57.80% | 67.59% |
| Math | 16.00% | 26.00% |
| Academic | 88.32% | 93.22% |
| Coding | 42.02% | 36.53% |
| Social Mobility | 62.92% | 53.79% |
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 | Gemini 3.1 Pro Preview (02/26) cost | DeepSeek V4 latency | Gemini 3.1 Pro Preview (02/26) latency |
|---|---|---|---|---|
| Vals Index | $1.31 | $2.18 | 32m09s | 9m52s |
| Harvey's Legal Agent Benchmark | $0.74 | $1.15 | 6m42s | 7m17s |
| Legal Research Bench | $0.68 | $1.10 | 17m05s | 4m02s |
| LegalBench | N/A | N/A | 3m04s | 10.06s |
| EMB | $0.80 | $3.77 | 13m42s | 11m49s |
| Finance Agent (v2) | $0.88 | $1.59 | 25m26s | 3m28s |
| Tax Agent Bench | $0.72 | $0.30 | 20m45s | 100.96s |
| TaxEval v2 | N/A | N/A | 3m10s | 41.93s |
| MedCode | N/A | N/A | 6m23s | 38.52s |
| MedScribe | N/A | N/A | 5m46s | 69.14s |
| ProofBench v1.1 | $0.03 | $0.78 | 7m03s | 14m49s |
| GPQA Diamond | N/A | N/A | 15m00s | 65.76s |
| MMLU Pro | N/A | N/A | 9m38s | 23.91s |
| Code Migration | $1.07 | $1.61 | 28m58s | 12m01s |
| LiveCodeBench | N/A | N/A | 11m56s | 88.89s |
| ProgramBench | N/A | N/A | 1h03m | 12m36s |
| SWE-bench | $0.44 | $0.78 | 10m35s | 5m12s |
| Terminal-Bench 4.0 | $3.38 | $4.11 | 1h28m | 18m44s |
| Vibe Code Bench v1.1 | $2.21 | $3.83 | 56m59s | 20m12s |
| Public Benefits Bench v1.1 | $0.46 | $0.38 | 15m22s | 3m06s |
Results available only for DeepSeek V4
- SkillsBench
Results available only for Gemini 3.1 Pro Preview (02/26)
- MortgageTax
- BioMysteryBench
- MysteryMechanism
- Terminal-Bench Science
- MMMU Pro
- SAGE
- IOI
- Vibe Code Bench 1-100