GPT-5.6 Luna vs Hy4 Preview: Benchmark Comparison
GPT-5.6 Luna has the higher score on 5 of 15 shared benchmarks; Hy4 Preview leads on 9.
The largest observed score gap is 15.00 pts on ProofBench v1.1 , where Hy4 Preview leads.
Reported ±1 standard-error ranges overlap on 9 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 | GPT-5.6 Luna | Hy4 Preview | Gap | Reported uncertainty |
|---|---|---|---|---|
| Legal Research Bench | 36.54% ±3.35 | 45.19% ±3.46 | 8.65 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 84.03% ±0.42 | 83.76% ±0.41 | 0.28 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 24.82% ±3.11 | 32.62% ±3.38 | 7.80 pts | Reported ±1 SE ranges do not overlap |
| MedCode | 42.39% ±2.27 | 43.25% ±2.13 | 0.86 pts | Reported ±1 SE ranges overlap |
| MedScribe | 84.39% ±2.58 | 83.60% ±2.06 | 0.79 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 60.00% ±4.92 | 75.00% ±4.35 | 15.00 pts | Reported ±1 SE ranges do not overlap |
| BioMysteryBench | 61.48% ±0.74 | 69.26% ±2.59 | 7.78 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench-Science | 0.00% ±0.00 | 1.43% ±1.43 | 1.43 pts | Reported ±1 SE ranges overlap |
| IOI | 61.78% ±11.61 | 59.33% ±4.60 | 2.45 pts | Reported ±1 SE ranges overlap |
| ProgramBench | 0.00% ±0.00 | 0.00% ±0.00 | 0.00 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 11.62% ±1.01 | 8.08% ±1.34 | 3.54 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 77.06% ±3.07 | 77.48% ±4.04 | 0.42 pts | Reported ±1 SE ranges overlap |
| CyberBench v1.1 | 73.63% ±5.56 | 65.36% ±5.55 | 8.27 pts | Reported ±1 SE ranges overlap |
| SRE Bench | 1.53% ±0.76 | 2.29% ±0.93 | 0.76 pts | Reported ±1 SE ranges overlap |
| Public Benefits Bench v1.1 | 61.16% ±1.27 | 68.61% ±1.21 | 7.44 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | GPT-5.6 Luna average | Hy4 Preview average |
|---|---|---|
| Legal | 60.29% | 64.47% |
| Finance | 24.82% | 32.62% |
| Healthcare | 63.39% | 63.42% |
| Math | 60.00% | 75.00% |
| Science | 30.74% | 35.34% |
| Coding | 37.61% | 36.22% |
| Cyber | 37.58% | 33.82% |
| Social Mobility | 61.16% | 68.61% |
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 | GPT-5.6 Luna cost | Hy4 Preview cost | GPT-5.6 Luna latency | Hy4 Preview latency |
|---|---|---|---|---|
| Legal Research Bench | $0.85 | $0.85 | 39m34s | 1h04m |
| LegalBench | N/A | N/A | 5.89s | 90.56s |
| Tax Agent Bench | $0.47 | $0.59 | 42m46s | 41m04s |
| MedCode | N/A | N/A | 81.28s | 6m39s |
| MedScribe | N/A | N/A | 116.89s | 5m50s |
| ProofBench v1.1 | $0.12 | $0.34 | 8m34s | 22m54s |
| BioMysteryBench | $0.09 | $0.36 | 18m31s | 23m30s |
| Terminal-Bench-Science | $0.58 | $2.14 | 2h03m | 3h48m |
| IOI | $0.58 | $1.35 | 1h10m | 1h01m |
| ProgramBench | $0.94 | $12.81 | 41m34s | 2h10m |
| Terminal-Bench 4.0 | $0.73 | $2.10 | 35m55s | 2h10m |
| Vibe Code Bench v1.1 | $0.73 | $2.18 | 25m52s | 41m33s |
| CyberBench v1.1 | $0.40 | $0.49 | 16m15s | 36m43s |
| SRE Bench | $3.44 | $4.90 | 1h42m | 1h40m |
| Public Benefits Bench v1.1 | $0.27 | $0.25 | 38m40s | 1h03m |
Results available only for GPT-5.6 Luna
- Vals Science Index
- MortgageTax
- TaxEval v2
- MysteryMechanism
- GPQA Diamond
- MMLU Pro
- MMMU Pro
- SAGE
- SkillsBench
- SWE-bench
- Vibe Code Bench 1-100
Results available only for Hy4 Preview
- Vals Finance Index
- Vals Index
- Vals Legal Index
- HLAB-Vals-1.1
- EMB v1.1
- Finance Agent v2.1
- Code Migration v1.1