Qwen 3.8 Max vs Step 5 Preview: Benchmark Comparison
Qwen 3.8 Max has the higher score on 9 of 11 shared benchmarks; Step 5 Preview leads on 2.
The largest observed score gap is 30.83 pts on IOI , where Qwen 3.8 Max leads.
Reported ±1 standard-error ranges overlap on 6 of 11 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 | Step 5 Preview | Gap | Reported uncertainty |
|---|---|---|---|---|
| Legal Research Bench | 47.60% ±3.47 | 40.38% ±3.41 | 7.21 pts | Reported ±1 SE ranges do not overlap |
| EMB v1.1 | 65.99% ±2.69 | 69.55% ±2.49 | 3.57 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 32.09% ±3.37 | 27.27% ±3.11 | 4.82 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 58.00% ±4.96 | 42.00% ±4.96 | 16.00 pts | Reported ±1 SE ranges do not overlap |
| MysteryMechanism | 23.87% ±2.87 | 18.47% ±2.61 | 5.41 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench-Science | 12.86% ±4.03 | 2.86% ±2.01 | 10.00 pts | Reported ±1 SE ranges do not overlap |
| IOI | 68.89% ±5.20 | 38.06% ±6.69 | 30.83 pts | Reported ±1 SE ranges do not overlap |
| SWE-bench Multimodal | 42.50% ±2.26 | 40.83% ±2.25 | 1.67 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 34.34% ±3.94 | 31.82% ±2.62 | 2.53 pts | Reported ±1 SE ranges overlap |
| Vibe Code Bench v1.1 | 64.70% ±5.36 | 70.40% ±4.27 | 5.70 pts | Reported ±1 SE ranges overlap |
| Public Benefits Bench v1.1 | 67.12% ±1.22 | 58.12% ±1.28 | 9.00 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Qwen 3.8 Max average | Step 5 Preview average |
|---|---|---|
| Legal | 47.60% | 40.38% |
| Finance | 49.04% | 48.41% |
| Math | 58.00% | 42.00% |
| Science | 18.37% | 10.66% |
| Coding | 52.61% | 45.28% |
| Social Mobility | 67.12% | 58.12% |
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 | Step 5 Preview cost | Qwen 3.8 Max latency | Step 5 Preview latency |
|---|---|---|---|---|
| Legal Research Bench | $2.49 | $0.84 | 1h01m | 36m08s |
| EMB v1.1 | $2.66 | $1.33 | 57m47s | 43m35s |
| Tax Agent Bench | $1.55 | $0.55 | 40m36s | 21m29s |
| ProofBench v1.1 | $1.50 | $0.71 | 31m08s | 38m34s |
| MysteryMechanism | $1.31 | $0.60 | 37m49s | 29m20s |
| Terminal-Bench-Science | $16.68 | $4.25 | 3h24m | 2h30m |
| IOI | $9.47 | $2.03 | 1h47m | 1h20m |
| SWE-bench Multimodal | $0.99 | $0.61 | 18m44s | 26m06s |
| Terminal-Bench 4.0 | $10.63 | $3.11 | 2h28m | 1h58m |
| Vibe Code Bench v1.1 | $8.24 | $5.86 | 2h49m | 1h31m |
| Public Benefits Bench v1.1 | $0.99 | $0.50 | 40m44s | 27m19s |
Results available only for Qwen 3.8 Max
- Vals RSI Index
- LegalBench
- MortgageTax
- TaxEval v2
- MedCode
- MedScribe
- GPQA Diamond
- MMLU Pro
- MMMU Pro
- SAGE
- LiveCodeBench
- ProgramBench
- SkillsBench
- SWE-bench
- Vibe Code Bench 1-100
- CyberBench v1.1
Results available only for Step 5 Preview
- Vals Finance Index
- Vals Science Index
- Finance Agent v2.1
- BioMysteryBench
- Code Migration v1.1
- SRE Bench