DeepSeek V4 vs Grok 4.5: Benchmark Comparison
DeepSeek V4 has the higher score on 4 of 15 shared benchmarks; Grok 4.5 leads on 10.
The largest observed score gap is 19.07 pts on Vibe Code Bench v1.1 , where Grok 4.5 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 | DeepSeek V4 | Grok 4.5 | Gap | Reported uncertainty |
|---|---|---|---|---|
| Legal Research Bench | 23.08% ±2.93 | 37.98% ±3.37 | 14.90 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 80.32% ±0.47 | 85.97% ±0.41 | 5.64 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 25.17% ±3.06 | 24.86% ±3.12 | 0.32 pts | Reported ±1 SE ranges overlap |
| TaxEval v2 | 72.08% ±0.88 | 71.67% ±0.89 | 0.41 pts | Reported ±1 SE ranges overlap |
| MedCode | 40.45% ±2.12 | 43.29% ±2.31 | 2.84 pts | Reported ±1 SE ranges overlap |
| MedScribe | 75.14% ±2.00 | 86.88% ±1.94 | 11.74 pts | Reported ±1 SE ranges do not overlap |
| ProofBench v1.1 | 16.00% ±3.69 | 31.00% ±4.65 | 15.00 pts | Reported ±1 SE ranges do not overlap |
| GPQA Diamond | 89.39% ±1.65 | 92.93% ±1.29 | 3.53 pts | Reported ±1 SE ranges do not overlap |
| MMLU Pro | 87.25% ±0.34 | 89.22% ±0.31 | 1.97 pts | Reported ±1 SE ranges do not overlap |
| LiveCodeBench | 87.48% ±0.95 | 87.35% ±0.96 | 0.13 pts | Reported ±1 SE ranges overlap |
| ProgramBench | 0.00% ±0.00 | 0.00% ±0.00 | 0.00 pts | Reported ±1 SE ranges overlap |
| SkillsBench | 51.27% ±4.60 | 66.03% ±4.60 | 14.76 pts | Reported ±1 SE ranges do not overlap |
| SWE-bench | 77.40% ±1.87 | 86.60% ±1.52 | 9.20 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 11.11% ±1.34 | 8.59% ±1.34 | 2.53 pts | Reported ±1 SE ranges overlap |
| Vibe Code Bench v1.1 | 49.93% ±4.77 | 69.00% ±4.48 | 19.07 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | DeepSeek V4 average | Grok 4.5 average |
|---|---|---|
| Legal | 51.70% | 61.97% |
| Finance | 48.62% | 48.26% |
| Healthcare | 57.80% | 65.09% |
| Math | 16.00% | 31.00% |
| Academic | 88.32% | 91.07% |
| Coding | 46.20% | 52.93% |
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 cost | Grok 4.5 cost | DeepSeek V4 latency | Grok 4.5 latency |
|---|---|---|---|---|
| Legal Research Bench | $0.68 | $1.09 | 17m05s | 15m54s |
| LegalBench | N/A | N/A | 3m04s | 67.88s |
| Tax Agent Bench | $0.72 | $0.47 | 20m45s | 4m44s |
| TaxEval v2 | N/A | N/A | 3m10s | 3m15s |
| MedCode | N/A | N/A | 6m23s | 59.64s |
| MedScribe | N/A | N/A | 5m46s | 11m45s |
| ProofBench v1.1 | $0.03 | $0.57 | 7m03s | 11m09s |
| GPQA Diamond | N/A | N/A | 15m00s | 3m38s |
| MMLU Pro | N/A | N/A | 9m38s | 5m58s |
| LiveCodeBench | N/A | N/A | 11m56s | 9m49s |
| ProgramBench | N/A | N/A | 1h03m | 18m59s |
| SkillsBench | $0.40 | $0.58 | 8m16s | 2m16s |
| SWE-bench | $0.44 | $0.54 | 10m35s | 3m20s |
| Terminal-Bench 4.0 | $3.45 | $5.52 | 1h26m | 33m43s |
| Vibe Code Bench v1.1 | $2.21 | $4.10 | 56m59s | 13m50s |
Results available only for DeepSeek V4
- Public Benefits Bench v1.1
Results available only for Grok 4.5
- MortgageTax
- MMMU Pro
- SAGE
- IOI
- CyberBench v1.1
- SRE Bench
- Time Horizon Index: KSP