Model comparison
Qwen 3.8 Max vs Claude Sonnet 4.6: Benchmark Comparison
Qwen 3.8 Max has the higher score on 12 of 19 shared benchmarks; Claude Sonnet 4.6 leads on 7.
The largest observed score gap is 15.93 pts on Code Migration , where Claude Sonnet 4.6 leads.
Reported ±1 standard-error ranges overlap on 7 of 19 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 | Qwen 3.8 Max | Claude Sonnet 4.6 | Gap | Reported uncertainty |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | 10.42% ±2.29 | 5.00% ±1.65 | 5.42 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 47.60% ±3.47 | 38.46% ±3.38 | 9.13 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 83.61% ±0.42 | 82.12% ±0.48 | 1.49 pts | Reported ±1 SE ranges do not overlap |
| EMB | 60.07% ±3.35 | 60.15% ±3.28 | 0.09 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 50.59% ±0.58 | 51.03% ±0.33 | 0.44 pts | Reported ±1 SE ranges overlap |
| MortgageTax | 63.99% ±0.94 | 67.73% ±0.92 | 3.74 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 65.96% ±3.18 | 64.85% ±3.11 | 1.11 pts | Reported ±1 SE ranges overlap |
| TaxEval v2 | 75.55% ±0.84 | 77.11% ±0.82 | 1.55 pts | Reported ±1 SE ranges overlap |
| GPQA Diamond | 93.69% ±1.22 | 85.61% ±2.41 | 8.08 pts | Reported ±1 SE ranges do not overlap |
| MMLU Pro | 88.60% ±0.31 | 87.34% ±0.43 | 1.26 pts | Reported ±1 SE ranges do not overlap |
| MMMU Pro | 88.03% ±0.78 | 83.58% ±0.89 | 4.45 pts | Reported ±1 SE ranges do not overlap |
| SAGE | 51.25% ±3.42 | 46.58% ±3.43 | 4.67 pts | Reported ±1 SE ranges overlap |
| Code Migration | 23.96% ±4.31 | 39.89% ±4.14 | 15.93 pts | Reported ±1 SE ranges do not overlap |
| LiveCodeBench | 87.85% ±0.95 | 82.09% ±1.06 | 5.76 pts | Reported ±1 SE ranges do not overlap |
| ProgramBench | 0.00% ±0.00 | 0.50% ±0.50 | 0.50 pts | Reported ±1 SE ranges overlap |
| SkillsBench | 42.01% ±4.43 | 49.05% ±4.64 | 7.04 pts | Reported ±1 SE ranges overlap |
| SWE-bench | 85.60% ±1.57 | 77.40% ±1.87 | 8.20 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 64.70% ±5.36 | 51.48% ±4.64 | 13.22 pts | Reported ±1 SE ranges do not overlap |
| Public Benefits Bench v1.1 | 67.12% ±1.22 | 62.45% ±1.26 | 4.67 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Qwen 3.8 Max average | Claude Sonnet 4.6 average |
|---|---|---|
| Legal | 47.21% | 41.86% |
| Finance | 63.23% | 64.17% |
| Academic | 90.11% | 85.51% |
| Education | 51.25% | 46.58% |
| Coding | 50.69% | 50.07% |
| Social Mobility | 67.12% | 62.45% |
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 | Qwen 3.8 Max cost | Claude Sonnet 4.6 cost | Qwen 3.8 Max latency | Claude Sonnet 4.6 latency |
|---|---|---|---|---|
| Harvey's Legal Agent Benchmark | $2.37 | $3.04 | 49m12s | 18m43s |
| Legal Research Bench | $2.49 | $2.30 | 1h01m | 16m30s |
| LegalBench | N/A | N/A | 36.37s | 21.63s |
| EMB | $2.66 | $7.23 | 57m47s | 54m16s |
| Finance Agent (v2) | $1.24 | $2.41 | 19m35s | 11m36s |
| MortgageTax | N/A | N/A | 47.50s | 29.74s |
| Tax Agent Bench | $1.55 | $1.19 | 40m36s | 11m47s |
| TaxEval v2 | N/A | N/A | 4m26s | 2m08s |
| GPQA Diamond | N/A | N/A | 4m27s | 6m18s |
| MMLU Pro | N/A | N/A | 63.58s | 55.41s |
| MMMU Pro | N/A | N/A | 43.85s | 2m12s |
| SAGE | N/A | N/A | 2m13s | 2m43s |
| Code Migration | $10.54 | $38.51 | 4h28m | 1h40m |
| LiveCodeBench | N/A | N/A | 5m38s | 4m55s |
| ProgramBench | N/A | N/A | 5h54m | 1h12m |
| SkillsBench | $0.52 | $1.38 | 15m48s | 14m20s |
| SWE-bench | $1.13 | $1.30 | 41m42s | 8m32s |
| Vibe Code Bench v1.1 | $8.24 | $5.91 | 2h49m | 26m12s |
| Public Benefits Bench v1.1 | $0.99 | $1.11 | 40m44s | 14m43s |
Results available only for Qwen 3.8 Max
- Vals Index
- Vals RSI Index
- MedCode
- MedScribe
- ProofBench v1.1
- MysteryMechanism
- Terminal-Bench Science
- IOI
- Terminal-Bench 4.0
- Vibe Code Bench 1-100
- CyberBench v1.1
Results available only for Claude Sonnet 4.6
None.