Claude Opus 5 vs GPT-5.6 Luna: Benchmark Comparison
Claude Opus 5 has the higher score on 26 of 29 shared benchmarks; GPT-5.6 Luna leads on 3.
The largest observed score gap is 41.92 pts on Terminal-Bench 4.0 , where Claude Opus 5 leads.
Reported ±1 standard-error ranges overlap on 6 of 29 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 5 | GPT-5.6 Luna | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Index | 63.67% ±0.96 | 51.69% ±1.06 | 11.99 pts | Reported ±1 SE ranges do not overlap |
| Harvey's Legal Agent Benchmark | 6.67% ±1.65 | 1.25% ±0.83 | 5.42 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 55.29% ±3.46 | 36.54% ±3.35 | 18.75 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 86.97% ±0.42 | 84.03% ±0.42 | 2.94 pts | Reported ±1 SE ranges do not overlap |
| EMB | 73.56% ±2.24 | 67.12% ±2.94 | 6.44 pts | Reported ±1 SE ranges do not overlap |
| Finance Agent (v2) | 58.63% ±0.08 | 55.04% ±0.31 | 3.59 pts | Reported ±1 SE ranges do not overlap |
| MortgageTax | 72.06% ±0.88 | 67.29% ±0.92 | 4.77 pts | Reported ±1 SE ranges do not overlap |
| Tax Agent Bench | 75.06% ±2.96 | 60.81% ±3.26 | 14.25 pts | Reported ±1 SE ranges do not overlap |
| TaxEval v2 | 75.14% ±0.83 | 76.17% ±0.84 | 1.02 pts | Reported ±1 SE ranges overlap |
| MedCode | 63.57% ±1.99 | 42.39% ±2.27 | 21.18 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 90.98% ±1.92 | 84.39% ±2.58 | 6.59 pts | Reported ±1 SE ranges do not overlap |
| ProofBench v1.1 | 99.00% ±1.00 | 60.00% ±4.92 | 39.00 pts | Reported ±1 SE ranges do not overlap |
| BioMysteryBench | 79.26% ±2.06 | 61.48% ±0.74 | 17.78 pts | Reported ±1 SE ranges do not overlap |
| MysteryMechanism | 37.39% ±3.25 | 14.41% ±2.36 | 22.97 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench Science | 27.14% ±5.35 | 0.00% ±0.00 | 27.14 pts | Reported ±1 SE ranges do not overlap |
| GPQA Diamond | 93.43% ±1.24 | 91.67% ±1.74 | 1.77 pts | Reported ±1 SE ranges overlap |
| MMLU Pro | 91.59% ±0.28 | 86.04% ±0.35 | 5.55 pts | Reported ±1 SE ranges do not overlap |
| MMMU Pro | 89.88% ±0.72 | 85.03% ±0.86 | 4.86 pts | Reported ±1 SE ranges do not overlap |
| SAGE | 49.43% ±3.29 | 44.22% ±3.33 | 5.21 pts | Reported ±1 SE ranges overlap |
| Code Migration | 57.47% ±4.37 | 44.55% ±4.24 | 12.92 pts | Reported ±1 SE ranges do not overlap |
| IOI | 84.33% ±9.96 | 61.78% ±11.61 | 22.55 pts | Reported ±1 SE ranges do not overlap |
| ProgramBench | 3.00% ±1.21 | 0.00% ±0.00 | 3.00 pts | Reported ±1 SE ranges do not overlap |
| SkillsBench | 60.44% ±4.58 | 60.45% ±4.75 | 0.01 pts | Reported ±1 SE ranges overlap |
| SWE-bench | 97.00% ±0.76 | 93.00% ±1.14 | 4.00 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 53.53% ±1.34 | 11.62% ±1.01 | 41.92 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench 1-100 | 28.53% ±4.33 | 22.59% ±4.00 | 5.94 pts | Reported ±1 SE ranges overlap |
| Vibe Code Bench v1.1 | 88.40% ±3.00 | 77.06% ±3.07 | 11.35 pts | Reported ±1 SE ranges do not overlap |
| CyberBench v1.1 | 65.36% ±5.55 | 73.63% ±5.56 | 8.27 pts | Reported ±1 SE ranges overlap |
| Public Benefits Bench v1.1 | 76.93% ±1.10 | 61.16% ±1.27 | 15.76 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Claude Opus 5 average | GPT-5.6 Luna average |
|---|---|---|
| Legal | 49.64% | 40.61% |
| Finance | 70.89% | 65.29% |
| Healthcare | 77.28% | 63.39% |
| Math | 99.00% | 60.00% |
| Science | 47.93% | 25.30% |
| Academic | 91.64% | 87.58% |
| Education | 49.43% | 44.22% |
| Coding | 59.09% | 46.38% |
| Cyber | 65.36% | 73.63% |
| Social Mobility | 76.93% | 61.16% |
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 5 cost | GPT-5.6 Luna cost | Claude Opus 5 latency | GPT-5.6 Luna latency |
|---|---|---|---|---|
| Vals Index | $19.31 | $0.82 | 57m41s | 29m27s |
| Harvey's Legal Agent Benchmark | $23.67 | $0.38 | 55m37s | 10m56s |
| Legal Research Bench | $6.76 | $0.85 | 31m01s | 39m34s |
| LegalBench | N/A | N/A | 4.42s | 5.89s |
| EMB | $6.07 | $0.37 | 25m59s | 11m00s |
| Finance Agent (v2) | $5.12 | $0.28 | 9m58s | 12m52s |
| MortgageTax | N/A | N/A | 7.98s | 22.84s |
| Tax Agent Bench | $5.14 | $0.47 | 16m48s | 42m46s |
| TaxEval v2 | N/A | N/A | 58.34s | 76.52s |
| MedCode | N/A | N/A | 24.50s | 81.28s |
| MedScribe | N/A | N/A | 76.56s | 116.89s |
| ProofBench v1.1 | $2.11 | $0.12 | 9m51s | 8m34s |
| BioMysteryBench | $3.29 | $0.09 | 36m36s | 18m31s |
| MysteryMechanism | $3.28 | $0.11 | 16m53s | 8m13s |
| Terminal-Bench Science | $32.54 | $0.58 | 2h12m | 2h03m |
| GPQA Diamond | N/A | N/A | 39.59s | 53.52s |
| MMLU Pro | N/A | N/A | 9.60s | 16.95s |
| MMMU Pro | N/A | N/A | 30.03s | 35.10s |
| SAGE | N/A | N/A | 56.17s | 51.88s |
| Code Migration | $60.51 | $1.88 | 2h51m | 58m35s |
| IOI | $16.48 | $0.58 | 56m42s | 1h10m |
| ProgramBench | $60.29 | $0.94 | 2h55m | 41m34s |
| SkillsBench | $2.51 | $0.21 | 8m25s | 5m20s |
| SWE-bench | $1.29 | $0.04 | 9m37s | 3m21s |
| Terminal-Bench 4.0 | $18.60 | $0.73 | 1h07m | 35m55s |
| Vibe Code Bench 1-100 | $41.49 | $4.91 | 1h22m | 2h03m |
| Vibe Code Bench v1.1 | $33.88 | $0.73 | 1h28m | 25m52s |
| CyberBench v1.1 | $2.87 | $0.40 | 19m25s | 16m15s |
| Public Benefits Bench v1.1 | $3.95 | $0.27 | 30m09s | 38m40s |
Results available only for Claude Opus 5
- Vals RSI Index
- LiveCodeBench
- SRE Bench
- CUA-bench
- Time Horizon Index: KSP
Results available only for GPT-5.6 Luna
None.