DeepSeek V4 Flash 0731 vs Gemini 3.5 Flash: Benchmark Comparison
DeepSeek V4 Flash 0731 has the higher score on 6 of 16 shared benchmarks; Gemini 3.5 Flash leads on 9.
The largest observed score gap is 26.05 pts on Vibe Code Bench v1.1 , where DeepSeek V4 Flash 0731 leads.
Reported ±1 standard-error ranges overlap on 7 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 Flash 0731 | Gemini 3.5 Flash | Gap | Reported uncertainty |
|---|---|---|---|---|
| Legal Research Bench | 30.29% ±3.19 | 30.77% ±3.21 | 0.48 pts | Reported ±1 SE ranges overlap |
| LegalBench | 77.71% ±0.50 | 83.60% ±0.85 | 5.89 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 28.07% ±3.16 | 21.65% ±2.96 | 6.42 pts | Reported ±1 SE ranges do not overlap |
| TaxEval v2 | 70.69% ±0.90 | 74.37% ±0.85 | 3.68 pts | Reported ±1 SE ranges do not overlap |
| MedCode | 41.41% ±2.15 | 55.83% ±2.11 | 14.41 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 80.36% ±1.97 | 76.57% ±1.92 | 3.79 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 56.00% ±4.99 | 31.00% ±4.65 | 25.00 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench-Science | 4.29% ±2.44 | 5.71% ±2.79 | 1.43 pts | Reported ±1 SE ranges overlap |
| GPQA Diamond | 89.90% ±1.69 | 92.68% ±1.46 | 2.78 pts | Reported ±1 SE ranges overlap |
| MMLU Pro | 86.21% ±0.34 | 89.52% ±0.31 | 3.31 pts | Reported ±1 SE ranges do not overlap |
| LiveCodeBench | 87.26% ±0.97 | 87.60% ±0.95 | 0.34 pts | Reported ±1 SE ranges overlap |
| ProgramBench | 0.00% ±0.00 | 0.00% ±0.00 | 0.00 pts | Reported ±1 SE ranges overlap |
| SkillsBench | 50.67% ±4.43 | 52.74% ±4.47 | 2.07 pts | Reported ±1 SE ranges overlap |
| SWE-bench | 88.80% ±1.41 | 78.80% ±1.83 | 10.00 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 18.69% ±3.54 | 6.06% ±1.51 | 12.63 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 74.74% ±3.44 | 48.68% ±4.73 | 26.05 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | DeepSeek V4 Flash 0731 average | Gemini 3.5 Flash average |
|---|---|---|
| Legal | 54.00% | 57.19% |
| Finance | 49.38% | 48.01% |
| Healthcare | 60.89% | 66.20% |
| Math | 56.00% | 31.00% |
| Science | 4.29% | 5.71% |
| Academic | 88.05% | 91.10% |
| Coding | 53.36% | 45.65% |
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 Flash 0731 cost | Gemini 3.5 Flash cost | DeepSeek V4 Flash 0731 latency | Gemini 3.5 Flash latency |
|---|---|---|---|---|
| Legal Research Bench | $0.28 | $1.31 | 21m07s | 4m38s |
| LegalBench | N/A | N/A | 3.95s | 34.58s |
| Tax Agent Bench | $0.16 | $0.86 | 5m37s | 3m53s |
| TaxEval v2 | N/A | N/A | 72.20s | 19.69s |
| MedCode | N/A | N/A | 2m07s | 25.29s |
| MedScribe | N/A | N/A | 77.55s | 57.08s |
| ProofBench v1.1 | $0.16 | $0.57 | 15m34s | 6m34s |
| Terminal-Bench-Science | $1.10 | $3.80 | 1h34m | 1h01m |
| GPQA Diamond | N/A | N/A | 79.96s | 27.66s |
| MMLU Pro | N/A | N/A | 17.19s | 26.26s |
| LiveCodeBench | N/A | N/A | 2m07s | 64.37s |
| ProgramBench | N/A | N/A | 5h39m | 18m25s |
| SkillsBench | $0.13 | $1.25 | 6m52s | 7m15s |
| SWE-bench | $0.01 | $0.95 | 3m22s | 4m14s |
| Terminal-Bench 4.0 | $0.77 | $5.98 | 1h46m | 1h11m |
| Vibe Code Bench v1.1 | $0.91 | $2.54 | 37m22s | 14m53s |
Results available only for DeepSeek V4 Flash 0731
- BioMysteryBench
- IOI
- CyberBench v1.1
Results available only for Gemini 3.5 Flash
- MortgageTax
- MMMU Pro
- Financial Coaching Bench
- Public Benefits Bench v1.1