DeepSeek V4 Pro 0813 vs Step 5 Preview: Benchmark Comparison
DeepSeek V4 Pro 0813 has the higher score on 7 of 12 shared benchmarks; Step 5 Preview leads on 5.
The largest observed score gap is 17.68 pts on Terminal-Bench 4.0 , where Step 5 Preview leads.
Reported ±1 standard-error ranges overlap on 8 of 12 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 Pro 0813 | Step 5 Preview | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Finance Index | 52.94% ±1.41 | 55.43% ±1.44 | 2.49 pts | Reported ±1 SE ranges overlap |
| Legal Research Bench | 40.87% ±3.42 | 40.38% ±3.41 | 0.48 pts | Reported ±1 SE ranges overlap |
| EMB v1.1 | 65.13% ±2.35 | 69.55% ±2.49 | 4.42 pts | Reported ±1 SE ranges overlap |
| Finance Agent v2.1 | 57.12% ±2.08 | 58.90% ±2.05 | 1.78 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 26.76% ±3.09 | 27.27% ±3.11 | 0.52 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 50.00% ±5.03 | 42.00% ±4.96 | 8.00 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench-Science | 4.29% ±2.44 | 2.86% ±2.01 | 1.43 pts | Reported ±1 SE ranges overlap |
| Code Migration v1.1 | 51.99% ±2.23 | 47.00% ±3.43 | 4.98 pts | Reported ±1 SE ranges overlap |
| IOI | 51.61% ±2.40 | 38.06% ±6.69 | 13.55 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 14.14% ±2.02 | 31.82% ±2.62 | 17.68 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 82.30% ±3.15 | 70.40% ±4.27 | 11.90 pts | Reported ±1 SE ranges do not overlap |
| SRE Bench | 1.91% ±0.85 | 0.38% ±0.38 | 1.53 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | DeepSeek V4 Pro 0813 average | Step 5 Preview average |
|---|---|---|
| Legal | 40.87% | 40.38% |
| Finance | 49.67% | 51.91% |
| Math | 50.00% | 42.00% |
| Science | 4.29% | 2.86% |
| Coding | 50.01% | 46.82% |
| Cyber | 1.91% | 0.38% |
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 Pro 0813 cost | Step 5 Preview cost | DeepSeek V4 Pro 0813 latency | Step 5 Preview latency |
|---|---|---|---|---|
| Vals Finance Index | $0.88 | $0.82 | 24m00s | 24m42s |
| Legal Research Bench | $1.09 | $0.84 | 42m07s | 36m08s |
| EMB v1.1 | $1.08 | $1.33 | 37m27s | 43m35s |
| Finance Agent v2.1 | $0.82 | $0.58 | 7m57s | 9m03s |
| Tax Agent Bench | $0.73 | $0.55 | 26m37s | 21m29s |
| ProofBench v1.1 | $0.06 | $0.71 | 12m12s | 38m34s |
| Terminal-Bench-Science | $4.63 | $4.25 | 2h16m | 2h30m |
| Code Migration v1.1 | $12.16 | $8.61 | 2h54m | 3h17m |
| IOI | $2.15 | $2.03 | 1h07m | 1h20m |
| Terminal-Bench 4.0 | $3.31 | $3.11 | 1h32m | 1h58m |
| Vibe Code Bench v1.1 | $0.36 | $5.86 | 1h06m | 1h31m |
| SRE Bench | $5.58 | $3.07 | 2h20m | 1h34m |
Results available only for DeepSeek V4 Pro 0813
- Vals Index
- Vals Legal Index
- HLAB-Vals-1.1
- LegalBench
- TaxEval v2
- MedCode
- MedScribe
- GPQA Diamond
- MMLU Pro
- LiveCodeBench
- ProgramBench
- SkillsBench
- SWE-bench
- Vibe Code Bench 1-100
- CyberBench v1.1
Results available only for Step 5 Preview
- Vals Science Index
- BioMysteryBench
- MysteryMechanism
- SWE-bench Multimodal
- Public Benefits Bench v1.1