Claude Haiku 5.5 vs GPT-5.6 Terra: Benchmark Comparison
Claude Haiku 5.5 has the higher score on 11 of 15 shared benchmarks; GPT-5.6 Terra leads on 3.
The largest observed score gap is 40.33 pts on IOI , where GPT-5.6 Terra leads.
Reported ±1 standard-error ranges overlap on 7 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 Haiku 5.5 | GPT-5.6 Terra | Gap | Reported uncertainty |
|---|---|---|---|---|
| Legal Research Bench | 43.27% ±3.44 | 41.35% ±3.42 | 1.92 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 31.92% ±3.35 | 30.04% ±3.31 | 1.89 pts | Reported ±1 SE ranges overlap |
| MedCode | 51.54% ±2.17 | 43.41% ±2.17 | 8.13 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 89.54% ±1.92 | 82.87% ±1.95 | 6.67 pts | Reported ±1 SE ranges do not overlap |
| ProofBench v1.1 | 89.00% ±3.15 | 74.00% ±4.41 | 15.00 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench-Science | 10.00% ±3.61 | 10.00% ±3.61 | 0.00 pts | Reported ±1 SE ranges overlap |
| SAGE | 45.78% ±3.42 | 47.00% ±3.40 | 1.22 pts | Reported ±1 SE ranges overlap |
| IOI | 47.28% ±10.72 | 87.61% ±6.53 | 40.33 pts | Reported ±1 SE ranges do not overlap |
| ProgramBench | 2.50% ±1.11 | 0.50% ±0.50 | 2.00 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 36.36% ±2.31 | 22.73% ±2.31 | 13.64 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench 1-100 | 20.41% ±3.79 | 14.82% ±3.12 | 5.58 pts | Reported ±1 SE ranges overlap |
| Vibe Code Bench v1.1 | 90.44% ±1.73 | 74.59% ±4.20 | 15.85 pts | Reported ±1 SE ranges do not overlap |
| CyberBench v1.1 | 75.48% ±5.21 | 72.08% ±5.41 | 3.39 pts | Reported ±1 SE ranges overlap |
| SRE Bench | 5.72% ±1.44 | 9.54% ±1.82 | 3.82 pts | Reported ±1 SE ranges do not overlap |
| Public Benefits Bench v1.1 | 63.19% ±1.25 | 62.38% ±1.26 | 0.81 pts | Reported ±1 SE ranges overlap |
Performance by category
| Category | Claude Haiku 5.5 average | GPT-5.6 Terra average |
|---|---|---|
| Legal | 43.27% | 41.35% |
| Finance | 31.92% | 30.04% |
| Healthcare | 70.54% | 63.14% |
| Math | 89.00% | 74.00% |
| Science | 10.00% | 10.00% |
| Education | 45.78% | 47.00% |
| Coding | 39.40% | 40.05% |
| Cyber | 40.60% | 40.81% |
| Social Mobility | 63.19% | 62.38% |
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 Haiku 5.5 cost | GPT-5.6 Terra cost | Claude Haiku 5.5 latency | GPT-5.6 Terra latency |
|---|---|---|---|---|
| Legal Research Bench | $1.63 | $7.91 | 1h22m | 1h10m |
| Tax Agent Bench | $1.16 | $4.83 | 49m34s | 1h01m |
| MedCode | N/A | N/A | 3m46s | 18.41s |
| MedScribe | N/A | N/A | 5m23s | 35.53s |
| ProofBench v1.1 | $0.17 | $1.27 | 6m58s | 12m57s |
| Terminal-Bench-Science | $4.52 | $5.19 | 1h34m | 2h04m |
| SAGE | N/A | N/A | 2m23s | 40.06s |
| IOI | $0.83 | $8.66 | 12m34s | 1h23m |
| ProgramBench | $6.83 | $5.38 | 1h13m | 33m36s |
| Terminal-Bench 4.0 | $3.12 | $5.60 | 51m56s | 31m29s |
| Vibe Code Bench 1-100 | $8.30 | $33.33 | 4h06m | 1h49m |
| Vibe Code Bench v1.1 | $6.07 | $7.89 | 34m36s | 19m12s |
| CyberBench v1.1 | $1.25 | $3.31 | 32m39s | 17m04s |
| SRE Bench | $3.14 | $30.86 | 57m56s | 3h17m |
| Public Benefits Bench v1.1 | $0.53 | $1.20 | 36m32s | 19m31s |
Results available only for Claude Haiku 5.5
- Vals Finance Index
- Vals Index
- Vals Legal Index
- Vals Science Index
- HLAB-Vals-1.1
- ClaimBench
- EMB v1.1
- Finance Agent v2.1
- BioMysteryBench
- MysteryMechanism
- Code Migration v1.1
- SWE-bench Multimodal
Results available only for GPT-5.6 Terra
- LegalBench
- MortgageTax
- TaxEval v2
- GPQA Diamond
- MMLU Pro
- MMMU Pro
- LiveCodeBench
- SkillsBench
- SWE-bench