Model comparison
Gemini 3.7 Flash vs GPT-6 Luna: Benchmark Comparison
Gemini 3.7 Flash has the higher score on 9 of 17 shared benchmarks; GPT-6 Luna leads on 8.
The largest observed score gap is 32.50 pts on CyberBench v1.1 , where GPT-6 Luna leads.
Reported ±1 standard-error ranges overlap on 11 of 17 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 | Gemini 3.7 Flash | GPT-6 Luna | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Index | 59.31% ±1.06 | 58.45% ±1.11 | 0.86 pts | Reported ±1 SE ranges overlap |
| Harvey's Legal Agent Benchmark | 8.75% ±2.15 | 2.92% ±1.17 | 5.83 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 34.62% ±3.31 | 30.29% ±3.19 | 4.33 pts | Reported ±1 SE ranges overlap |
| EMB | 71.33% ±2.26 | 68.52% ±2.81 | 2.81 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 59.04% ±0.27 | 49.87% ±0.23 | 9.17 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 57.66% ±3.31 | 58.86% ±3.21 | 1.20 pts | Reported ±1 SE ranges overlap |
| MedCode | 53.39% ±2.12 | 44.69% ±2.30 | 8.70 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 83.94% ±2.00 | 83.71% ±1.95 | 0.23 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 58.00% ±4.96 | 64.00% ±4.82 | 6.00 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench Science | 5.71% ±2.79 | 10.00% ±3.61 | 4.29 pts | Reported ±1 SE ranges overlap |
| SAGE | 49.23% ±3.38 | 48.09% ±3.38 | 1.14 pts | Reported ±1 SE ranges overlap |
| Code Migration | 34.80% ±4.22 | 42.55% ±4.42 | 7.75 pts | Reported ±1 SE ranges overlap |
| IOI | 67.83% ±3.61 | 55.56% ±8.87 | 12.28 pts | Reported ±1 SE ranges overlap |
| ProgramBench | 0.00% ±0.00 | 0.50% ±0.50 | 0.50 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 6.06% ±0.88 | 9.60% ±1.01 | 3.54 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 70.39% ±4.84 | 81.65% ±3.38 | 11.25 pts | Reported ±1 SE ranges do not overlap |
| CyberBench v1.1 | 43.75% ±2.21 | 76.25% ±5.29 | 32.50 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Gemini 3.7 Flash average | GPT-6 Luna average |
|---|---|---|
| Index | 59.31% | 58.45% |
| Legal | 21.68% | 16.60% |
| Finance | 62.68% | 59.08% |
| Healthcare | 68.67% | 64.20% |
| Math | 58.00% | 64.00% |
| Science | 5.71% | 10.00% |
| Education | 49.23% | 48.09% |
| Coding | 35.82% | 37.97% |
| Beta | 43.75% | 76.25% |
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 | Gemini 3.7 Flash cost | GPT-6 Luna cost | Gemini 3.7 Flash latency | GPT-6 Luna latency |
|---|---|---|---|---|
| Vals Index | $4.17 | $0.42 | 43m13s | 26m30s |
| Harvey's Legal Agent Benchmark | $2.56 | $0.30 | 7m32s | 17m32s |
| Legal Research Bench | $0.76 | $0.44 | 4m49s | 42m28s |
| EMB | $5.69 | $0.14 | 10m51s | 17m21s |
| Finance Agent (v2) | $1.48 | $0.12 | 3m01s | 14m58s |
| Tax Agent Bench | $0.48 | $0.19 | 87.24s | 33m16s |
| MedCode | N/A | N/A | 9.16s | 108.77s |
| MedScribe | N/A | N/A | 16.82s | 2m50s |
| ProofBench v1.1 | $0.56 | $0.04 | 7m04s | 7m31s |
| Terminal-Bench Science | $13.22 | $0.34 | 1h08m | 1h51m |
| SAGE | N/A | N/A | 9.79s | 91.07s |
| Code Migration | $21.46 | $0.60 | 2h39m | 50m23s |
| IOI | $3.40 | $0.20 | 21m09s | 35m47s |
| ProgramBench | $7.13 | $0.18 | 21m30s | 28m55s |
| Terminal-Bench 4.0 | $14.07 | $0.29 | 1h59m | 1h12m |
| Vibe Code Bench v1.1 | $4.83 | $1.35 | 23m43s | 37m00s |
| CyberBench v1.1 | $2.15 | $0.13 | 7m34s | 15m50s |
Results available only for Gemini 3.7 Flash
- Vals RSI Index
- LegalBench
- MortgageTax
- TaxEval v2
- GPQA Diamond
- MMLU Pro
- MMMU Pro
- LiveCodeBench
- SkillsBench
- SWE-bench
- SRE Bench
Results available only for GPT-6 Luna
- BioMysteryBench
- MysteryMechanism
- Public Benefits Bench v1.1