Claude Opus 4.8 vs GPT-6 Luna: Benchmark Comparison
Claude Opus 4.8 has the higher score on 14 of 15 shared benchmarks; GPT-6 Luna leads on 0.
The largest observed score gap is 13.46 pts on Legal Research Bench , where Claude Opus 4.8 leads.
Reported ±1 standard-error ranges overlap on 8 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 | Claude Opus 4.8 | GPT-6 Luna | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Index | 55.10% ±1.01 | 51.22% ±1.07 | 3.88 pts | Reported ±1 SE ranges do not overlap |
| Harvey's Legal Agent Benchmark | 9.58% ±2.15 | 2.92% ±1.36 | 6.67 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 43.75% ±3.45 | 30.29% ±3.19 | 13.46 pts | Reported ±1 SE ranges do not overlap |
| EMB | 69.37% ±2.61 | 68.52% ±2.81 | 0.86 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 53.92% ±0.16 | 49.87% ±0.23 | 4.05 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 64.91% ±3.17 | 58.86% ±3.21 | 6.04 pts | Reported ±1 SE ranges overlap |
| MedCode | 53.22% ±2.17 | 44.69% ±2.30 | 8.53 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 85.75% ±1.93 | 83.71% ±1.95 | 2.05 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench Science | 4.29% ±2.44 | 4.29% ±2.44 | 0.00 pts | Reported ±1 SE ranges overlap |
| SAGE | 54.79% ±3.35 | 48.09% ±3.38 | 6.70 pts | Reported ±1 SE ranges overlap |
| Code Migration | 47.25% ±4.18 | 42.55% ±4.42 | 4.70 pts | Reported ±1 SE ranges overlap |
| ProgramBench | 1.00% ±0.70 | 0.50% ±0.50 | 0.50 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 23.23% ±1.34 | 13.64% ±1.51 | 9.60 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 82.72% ±3.08 | 81.65% ±3.38 | 1.08 pts | Reported ±1 SE ranges overlap |
| Public Benefits Bench v1.1 | 68.13% ±1.21 | 57.65% ±1.28 | 10.49 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Claude Opus 4.8 average | GPT-6 Luna average |
|---|---|---|
| Legal | 26.67% | 16.60% |
| Finance | 62.73% | 59.08% |
| Healthcare | 69.49% | 64.20% |
| Science | 4.29% | 4.29% |
| Education | 54.79% | 48.09% |
| Coding | 38.55% | 34.58% |
| Social Mobility | 68.13% | 57.65% |
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 | Claude Opus 4.8 cost | GPT-6 Luna cost | Claude Opus 4.8 latency | GPT-6 Luna latency |
|---|---|---|---|---|
| Vals Index | $13.14 | $0.43 | 39m48s | 30m34s |
| Harvey's Legal Agent Benchmark | $10.22 | $0.30 | 24m08s | 17m32s |
| Legal Research Bench | $2.82 | $0.44 | 11m54s | 42m28s |
| EMB | $12.06 | $0.14 | 43m52s | 17m21s |
| Finance Agent (v2) | $4.22 | $0.12 | 9m07s | 14m58s |
| Tax Agent Bench | $1.69 | $0.19 | 7m59s | 33m16s |
| MedCode | N/A | N/A | 105.92s | 108.77s |
| MedScribe | N/A | N/A | 82.36s | 2m50s |
| Terminal-Bench Science | $23.14 | $0.24 | 1h55m | 1h16m |
| SAGE | N/A | N/A | 3m06s | 91.07s |
| Code Migration | $30.51 | $0.60 | 1h15m | 50m23s |
| ProgramBench | $31.27 | $0.18 | 1h32m | 28m55s |
| Terminal-Bench 4.0 | $17.14 | $0.35 | 1h11m | 32m46s |
| Vibe Code Bench v1.1 | $26.88 | $1.35 | 1h16m | 37m00s |
| Public Benefits Bench v1.1 | $1.89 | $0.22 | 12m48s | 52m59s |
Results available only for Claude Opus 4.8
- Vals RSI Index
- LegalBench
- MortgageTax
- TaxEval v2
- GPQA Diamond
- MMLU Pro
- MMMU Pro
- LiveCodeBench
- SkillsBench
- SWE-bench
- Time Horizon Index: KSP
Results available only for GPT-6 Luna
- ProofBench v1.1
- BioMysteryBench
- MysteryMechanism
- IOI
- CyberBench v1.1