Model comparison
Claude Sonnet 4.6 vs DeepSeek V4 Pro 0813: Benchmark Comparison
Claude Sonnet 4.6 has the higher score on 6 of 15 shared benchmarks; DeepSeek V4 Pro 0813 leads on 9.
The largest observed score gap is 30.82 pts on Vibe Code Bench v1.1 , where DeepSeek V4 Pro 0813 leads.
Reported ±1 standard-error ranges overlap on 8 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 Pro 0813 | Gap | Reported uncertainty |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | 5.00% ±1.65 | 7.50% ±2.15 | 2.50 pts | Reported ±1 SE ranges overlap |
| Legal Research Bench | 38.46% ±3.38 | 40.87% ±3.42 | 2.40 pts | Reported ±1 SE ranges overlap |
| LegalBench | 82.12% ±0.48 | 82.36% ±0.44 | 0.24 pts | Reported ±1 SE ranges overlap |
| EMB | 60.15% ±3.28 | 52.80% ±3.06 | 7.35 pts | Reported ±1 SE ranges do not overlap |
| Finance Agent (v2) | 51.03% ±0.33 | 50.39% ±0.25 | 0.64 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 64.85% ±3.11 | 58.66% ±3.14 | 6.19 pts | Reported ±1 SE ranges overlap |
| TaxEval v2 | 77.11% ±0.82 | 73.06% ±0.87 | 4.05 pts | Reported ±1 SE ranges do not overlap |
| GPQA Diamond | 85.61% ±2.41 | 92.42% ±2.02 | 6.82 pts | Reported ±1 SE ranges do not overlap |
| MMLU Pro | 87.34% ±0.43 | 86.97% ±0.34 | 0.37 pts | Reported ±1 SE ranges overlap |
| Code Migration | 39.89% ±4.14 | 41.54% ±4.30 | 1.65 pts | Reported ±1 SE ranges overlap |
| LiveCodeBench | 82.09% ±1.06 | 87.53% ±0.96 | 5.43 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 | 53.83% ±4.49 | 4.78 pts | Reported ±1 SE ranges overlap |
| SWE-bench | 77.40% ±1.87 | 96.40% ±0.83 | 19.00 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 51.48% ±4.64 | 82.30% ±3.15 | 30.82 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Claude Sonnet 4.6 average | DeepSeek V4 Pro 0813 average |
|---|---|---|
| Legal | 41.86% | 43.58% |
| Finance | 63.29% | 58.73% |
| Academic | 86.47% | 89.70% |
| Coding | 50.07% | 60.27% |
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 Pro 0813 cost | Claude Sonnet 4.6 latency | DeepSeek V4 Pro 0813 latency |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | $3.04 | $0.17 | 18m43s | 21m53s |
| Legal Research Bench | $2.30 | $1.09 | 16m30s | 42m07s |
| LegalBench | N/A | N/A | 21.63s | 20.05s |
| EMB | $7.23 | $1.08 | 54m16s | 37m27s |
| Finance Agent (v2) | $2.41 | $0.88 | 11m36s | 16m15s |
| Tax Agent Bench | $1.19 | $0.73 | 11m47s | 26m37s |
| TaxEval v2 | N/A | N/A | 2m08s | 3m46s |
| GPQA Diamond | N/A | N/A | 6m18s | 4m39s |
| MMLU Pro | N/A | N/A | 55.41s | 59.62s |
| Code Migration | $38.51 | $18.58 | 1h40m | 3h32m |
| LiveCodeBench | N/A | N/A | 4m55s | 4m33s |
| ProgramBench | N/A | N/A | 1h12m | 1h23m |
| SkillsBench | $1.38 | $0.37 | 14m20s | 11m15s |
| SWE-bench | $1.30 | $0.10 | 8m32s | 4m00s |
| Vibe Code Bench v1.1 | $5.91 | $0.36 | 26m12s | 1h06m |
Results available only for Claude Sonnet 4.6
- MortgageTax
- MMMU Pro
- SAGE
- Public Benefits Bench v1.1
Results available only for DeepSeek V4 Pro 0813
- Vals Index
- MedCode
- MedScribe
- ProofBench v1.1
- Terminal-Bench Science
- IOI
- Terminal-Bench 4.0
- Vibe Code Bench 1-100
- CyberBench v1.1