Model comparison
Claude Sonnet 4.6 vs DeepSeek V4 Flash 0731: Benchmark Comparison
Claude Sonnet 4.6 has the higher score on 9 of 15 shared benchmarks; DeepSeek V4 Flash 0731 leads on 6.
The largest observed score gap is 23.26 pts on Vibe Code Bench v1.1 , where DeepSeek V4 Flash 0731 leads.
Reported ±1 standard-error ranges overlap on 6 of 15 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 | Claude Sonnet 4.6 | DeepSeek V4 Flash 0731 | Gap | Reported uncertainty |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | 5.00% ±1.65 | 8.33% ±2.15 | 3.33 pts | Reported ±1 SE ranges overlap |
| Legal Research Bench | 38.46% ±3.38 | 30.29% ±3.19 | 8.17 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 82.12% ±0.48 | 77.71% ±0.50 | 4.41 pts | Reported ±1 SE ranges do not overlap |
| EMB | 60.15% ±3.28 | 56.98% ±3.39 | 3.18 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 51.03% ±0.33 | 49.52% ±0.47 | 1.52 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 64.85% ±3.11 | 62.56% ±3.14 | 2.29 pts | Reported ±1 SE ranges overlap |
| TaxEval v2 | 77.11% ±0.82 | 70.69% ±0.90 | 6.42 pts | Reported ±1 SE ranges do not overlap |
| GPQA Diamond | 85.61% ±2.41 | 89.90% ±1.69 | 4.29 pts | Reported ±1 SE ranges do not overlap |
| MMLU Pro | 87.34% ±0.43 | 86.21% ±0.34 | 1.13 pts | Reported ±1 SE ranges do not overlap |
| Code Migration | 39.89% ±4.14 | 38.63% ±4.38 | 1.26 pts | Reported ±1 SE ranges overlap |
| LiveCodeBench | 82.09% ±1.06 | 87.26% ±0.97 | 5.17 pts | Reported ±1 SE ranges do not overlap |
| ProgramBench | 0.50% ±0.50 | 0.00% ±0.00 | 0.50 pts | Reported ±1 SE ranges overlap |
| SkillsBench | 49.05% ±4.64 | 50.67% ±4.43 | 1.62 pts | Reported ±1 SE ranges overlap |
| SWE-bench | 77.40% ±1.87 | 88.80% ±1.41 | 11.40 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 51.48% ±4.64 | 74.74% ±3.44 | 23.26 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Claude Sonnet 4.6 average | DeepSeek V4 Flash 0731 average |
|---|---|---|
| Legal | 41.86% | 38.78% |
| Finance | 63.29% | 59.94% |
| Academic | 86.47% | 88.05% |
| Coding | 50.07% | 56.68% |
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 | Claude Sonnet 4.6 cost | DeepSeek V4 Flash 0731 cost | Claude Sonnet 4.6 latency | DeepSeek V4 Flash 0731 latency |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | $3.04 | $0.06 | 18m43s | 16m51s |
| Legal Research Bench | $2.30 | $0.28 | 16m30s | 21m07s |
| LegalBench | N/A | N/A | 21.63s | 3.95s |
| EMB | $7.23 | $0.38 | 54m16s | 26m14s |
| Finance Agent (v2) | $2.41 | $0.28 | 11m36s | 8m10s |
| Tax Agent Bench | $1.19 | $0.16 | 11m47s | 5m37s |
| TaxEval v2 | N/A | N/A | 2m08s | 72.20s |
| GPQA Diamond | N/A | N/A | 6m18s | 79.96s |
| MMLU Pro | N/A | N/A | 55.41s | 17.19s |
| Code Migration | $38.51 | $3.61 | 1h40m | 2h08m |
| LiveCodeBench | N/A | N/A | 4m55s | 2m07s |
| ProgramBench | N/A | N/A | 1h12m | 5h39m |
| SkillsBench | $1.38 | $0.13 | 14m20s | 6m52s |
| SWE-bench | $1.30 | $0.01 | 8m32s | 3m22s |
| Vibe Code Bench v1.1 | $5.91 | $0.91 | 26m12s | 37m22s |
Results available only for Claude Sonnet 4.6
- MortgageTax
- MMMU Pro
- SAGE
- Public Benefits Bench v1.1
Results available only for DeepSeek V4 Flash 0731
- Vals Index
- MedCode
- MedScribe
- ProofBench v1.1
- BioMysteryBench
- Terminal-Bench Science
- IOI
- Terminal-Bench 4.0
- CyberBench v1.1