Claude Fable 5 vs GPT-6 Astra: Benchmark Comparison
Claude Fable 5 has the higher score on 9 of 16 shared benchmarks; GPT-6 Astra leads on 7.
The largest observed score gap is 47.14 pts on Terminal-Bench Science , where GPT-6 Astra leads.
Reported ±1 standard-error ranges overlap on 6 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 Fable 5 | GPT-6 Astra | Gap | Reported uncertainty |
|---|---|---|---|---|
| Vals Index | 61.39% ±1.00 | 63.13% ±1.19 | 1.74 pts | Reported ±1 SE ranges overlap |
| Vals RSI Index | 24.32% | 27.07% | 2.75 pts | Uncertainty comparison unavailable |
| Harvey's Legal Agent Benchmark | 11.25% ±2.15 | 5.42% ±1.17 | 5.83 pts | Reported ±1 SE ranges do not overlap |
| Legal Research Bench | 49.52% ±3.48 | 39.42% ±3.40 | 10.10 pts | Reported ±1 SE ranges do not overlap |
| EMB | 73.67% ±2.47 | 71.70% ±2.48 | 1.97 pts | Reported ±1 SE ranges overlap |
| Finance Agent (v2) | 56.31% ±0.84 | 53.54% ±2.08 | 2.77 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 69.82% ±3.12 | 63.34% ±3.15 | 6.49 pts | Reported ±1 SE ranges do not overlap |
| MedCode | 56.07% ±2.20 | 48.49% ±2.13 | 7.58 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 88.52% ±1.95 | 87.91% ±1.94 | 0.61 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 95.00% ±2.19 | 99.00% ±1.00 | 4.00 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench Science | 15.71% ±4.38 | 62.86% ±5.82 | 47.14 pts | Reported ±1 SE ranges do not overlap |
| SAGE | 51.89% ±3.40 | 46.37% ±3.44 | 5.52 pts | Reported ±1 SE ranges overlap |
| Code Migration | 55.06% ±4.61 | 67.74% ±4.22 | 12.67 pts | Reported ±1 SE ranges do not overlap |
| ProgramBench | 2.00% ±0.99 | 5.50% ±1.62 | 3.50 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench 4.0 | 41.41% ±1.34 | 59.60% ±4.40 | 18.18 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench v1.1 | 90.35% ±2.10 | 89.59% ±2.17 | 0.76 pts | Reported ±1 SE ranges overlap |
Performance by category
| Category | Claude Fable 5 average | GPT-6 Astra average |
|---|---|---|
| Legal | 30.38% | 22.42% |
| Finance | 66.60% | 62.86% |
| Healthcare | 72.30% | 68.20% |
| Math | 95.00% | 99.00% |
| Science | 15.71% | 62.86% |
| Education | 51.89% | 46.37% |
| Coding | 47.21% | 55.61% |
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 | Claude Fable 5 cost | GPT-6 Astra cost | Claude Fable 5 latency | GPT-6 Astra latency |
|---|---|---|---|---|
| Vals Index | $29.59 | $18.46 | 42m24s | 27m48s |
| Vals RSI Index | $491.22 | $759.51 | 90h00m | 90h00m |
| Harvey's Legal Agent Benchmark | $19.23 | $26.16 | 26m53s | 24m49s |
| Legal Research Bench | $9.79 | $10.58 | 22m18s | 24m51s |
| EMB | $12.35 | $5.83 | 26m34s | 13m24s |
| Finance Agent (v2) | $8.06 | $6.82 | 10m11s | 11m33s |
| Tax Agent Bench | $6.70 | $5.80 | 14m27s | 15m01s |
| MedCode | N/A | N/A | 91.44s | 102.86s |
| MedScribe | N/A | N/A | 119.47s | 2m34s |
| ProofBench v1.1 | $5.45 | $1.71 | 13m43s | 5m08s |
| Terminal-Bench Science | $51.99 | $20.80 | 1h59m | 1h49m |
| SAGE | N/A | N/A | 116.92s | 41.63s |
| Code Migration | $112.10 | $44.36 | 1h49m | 58m30s |
| ProgramBench | $75.68 | $11.57 | 2h37m | 27m02s |
| Terminal-Bench 4.0 | $30.34 | $9.58 | 1h08m | 35m42s |
| Vibe Code Bench v1.1 | $41.71 | $38.51 | 1h02m | 39m04s |
Results available only for Claude Fable 5
- LegalBench
- MortgageTax
- TaxEval v2
- GPQA Diamond
- MMLU Pro
- MMMU Pro
- LiveCodeBench
- SWE-bench
- Public Benefits Bench v1.1
Results available only for GPT-6 Astra
- BioMysteryBench
- MysteryMechanism
- IOI
- Vibe Code Bench 1-100
- CyberBench v1.1
- SRE Bench
- CUA-bench
- Time Horizon Index: KSP