DeepSeek V4.1 Flash vs GPT-6 Luna: Benchmark Comparison
DeepSeek V4.1 Flash has the higher score on 16 of 24 shared benchmarks; GPT-6 Luna leads on 7.
The largest observed score gap is 15.28 pts on IOI , where GPT-6 Luna leads.
Reported ±1 standard-error ranges overlap on 10 of 24 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 V4.1 Flash | GPT-6 Luna | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Finance Index | 57.08% ±1.45 | 53.05% ±1.43 | 4.03 pts | Reported ±1 SE ranges do not overlap |
| Vals Science Index | 29.65% ±0.99 | 28.38% ±1.48 | 1.27 pts | Reported ±1 SE ranges overlap |
| Immigration | 65.64% ±1.64 | 62.53% ±1.57 | 3.11 pts | Reported ±1 SE ranges overlap |
| Legal Research Bench | 41.35% ±3.42 | 30.29% ±3.19 | 11.06 pts | Reported ±1 SE ranges do not overlap |
| EMB v1.1 | 67.76% ±2.56 | 69.95% ±2.54 | 2.20 pts | Reported ±1 SE ranges overlap |
| Finance Agent v2.1 | 65.43% ±2.02 | 56.84% ±2.12 | 8.59 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 26.63% ±3.14 | 19.95% ±2.87 | 6.68 pts | Reported ±1 SE ranges do not overlap |
| MedCode | 41.17% ±2.04 | 44.69% ±2.30 | 3.51 pts | Reported ±1 SE ranges overlap |
| MedScribe | 85.50% ±1.92 | 83.71% ±1.95 | 1.79 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 54.00% ±5.01 | 64.00% ±4.82 | 10.00 pts | Reported ±1 SE ranges do not overlap |
| BioMysteryBench | 67.78% ±1.11 | 61.48% ±2.59 | 6.30 pts | Reported ±1 SE ranges do not overlap |
| MysteryMechanism | 21.17% ±2.75 | 19.37% ±2.66 | 1.80 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench-Science | 0.00% ±0.00 | 4.29% ±2.44 | 4.29 pts | Reported ±1 SE ranges do not overlap |
| SAGE | 85.38% ±1.96 | 83.40% ±2.30 | 1.98 pts | Reported ±1 SE ranges overlap |
| Code Migration v1.1 | 58.94% ±2.66 | 46.67% ±3.61 | 12.27 pts | Reported ±1 SE ranges do not overlap |
| IOI | 40.28% ±2.56 | 55.56% ±8.87 | 15.28 pts | Reported ±1 SE ranges do not overlap |
| ProgramBench | 0.50% ±0.50 | 0.50% ±0.50 | 0.00 pts | Reported ±1 SE ranges overlap |
| SWE-bench Multimodal | 46.46% ±2.28 | 37.50% ±2.21 | 8.96 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 19.70% ±1.75 | 13.64% ±1.51 | 6.06 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 84.74% ±2.89 | 81.65% ±3.38 | 3.09 pts | Reported ±1 SE ranges overlap |
| CyberBench v1.1 | 73.69% ±5.48 | 76.25% ±5.29 | 2.56 pts | Reported ±1 SE ranges overlap |
| SRE Bench | 0.76% ±0.54 | 2.67% ±1.00 | 1.91 pts | Reported ±1 SE ranges do not overlap |
| Financial Coaching Bench | 72.83% ±1.57 | 65.24% ±1.68 | 7.59 pts | Reported ±1 SE ranges do not overlap |
| Public Benefits Bench v1.1 | 64.28% ±1.25 | 57.65% ±1.28 | 6.63 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | DeepSeek V4.1 Flash average | GPT-6 Luna average |
|---|---|---|
| Legal | 53.49% | 46.41% |
| Finance | 53.27% | 48.91% |
| Healthcare | 63.34% | 64.20% |
| Math | 54.00% | 64.00% |
| Science | 29.65% | 28.38% |
| Education | 85.38% | 83.40% |
| Coding | 41.77% | 39.25% |
| Cyber | 37.23% | 39.46% |
| Social Mobility | 68.55% | 61.44% |
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 V4.1 Flash cost | GPT-6 Luna cost | DeepSeek V4.1 Flash latency | GPT-6 Luna latency |
|---|---|---|---|---|
| Vals Finance Index | $0.25 | $0.15 | 11m37s | 22m01s |
| Vals Science Index | $0.29 | $0.11 | 1h07m | 30m15s |
| Immigration | $0.05 | $0.01 | 3m45s | 2m31s |
| Legal Research Bench | $0.25 | $0.44 | 11m42s | 42m28s |
| EMB v1.1 | $0.23 | $0.14 | 13m00s | 17m21s |
| Finance Agent v2.1 | $0.37 | $0.12 | 16m00s | 15m27s |
| Tax Agent Bench | $0.13 | $0.19 | 5m49s | 33m16s |
| MedCode | N/A | N/A | 35.13s | 108.77s |
| MedScribe | N/A | N/A | 53.83s | 2m50s |
| ProofBench v1.1 | $0.13 | $0.04 | 9m07s | 7m31s |
| BioMysteryBench | $0.23 | $0.05 | 14m50s | 8m44s |
| MysteryMechanism | $0.16 | $0.03 | 8m04s | 6m01s |
| Terminal-Bench-Science | $0.47 | $0.24 | 2h57m | 1h16m |
| SAGE | $0.01 | $0.01 | 33.11s | 47.63s |
| Code Migration v1.1 | $1.09 | $0.83 | 1h54m | 1h00m |
| IOI | $0.24 | $0.20 | 18m30s | 35m47s |
| ProgramBench | $0.90 | $0.18 | 1h02m | 28m55s |
| SWE-bench Multimodal | $0.23 | $0.03 | 19m24s | 5m16s |
| Terminal-Bench 4.0 | $0.50 | $0.35 | 48m22s | 32m46s |
| Vibe Code Bench v1.1 | $0.41 | $1.35 | 15m28s | 37m00s |
| CyberBench v1.1 | $0.11 | $0.13 | 13m50s | 15m50s |
| SRE Bench | $0.55 | $1.55 | 1h01m | 1h21m |
| Financial Coaching Bench | $0.02 | $0.02 | 6m08s | 5m49s |
| Public Benefits Bench v1.1 | $0.07 | $0.22 | 9m59s | 52m59s |
Results available only for DeepSeek V4.1 Flash
- Vals Inference Provider Index
- LegalBench
- SkillsBench
- Access to Justice Benchmark
Results available only for GPT-6 Luna
- Vals Index
- Vals Legal Index
- HLAB-Vals-1.1
- ClaimBench