Claude Fable 5.1 vs Claude Opus 4.8
Claude Fable 5.1 Verdict
- Claude Fable 5.1 (xhigh) leads on quality: 70.3 vs 64.7.
- Claude Fable 5.1 (xhigh) is stronger in coding, composite, knowledge, long context, reasoning.
- Claude Opus 4.8 is stronger in agents & tools, human preference, instruction following, multimodal.
- Claude Opus 4.8 is 2.0× cheaper ($10.00 vs $20.00 per 1M blended).
| Metric | Claude Fable 5.1 (xhigh) | Claude Opus 4.8 |
|---|---|---|
| BenchLeader Index | 70.3 | 64.7 |
| Coding score | 71.3 | 65.7 |
| Composite score | 95.0 | 81.9 |
| Knowledge score | 84.1 | 66.3 |
| Long context score | 68.1 | 65.3 |
| Reasoning score | 76.4 | 64.2 |
| Agents & tools score | – | 65.0 |
| Human preference score | – | 65.6 |
| Instruction following score | – | 61.6 |
| Multimodal score | – | 64.4 |
| Blended price $/M | $20.00 | $10.00 |
| Output speed | 58 tok/s | 58 tok/s |
| Time to first answer | 132.0 s | 29.5 s |
| Context window | 1M | 1M |
| Humanity's Last Exam | 46.5% | – |
| SimpleBench | – | 64.8% |
| SciCode | 60.1% | – |
| Remote Labor Index | – | 8.3% |
| FrontierCode | – | 46.5% |
| GSO-Bench | – | 47.1% |
| Epoch Capabilities Index | – | 158.3 |
| LMArena Text | – | 1473 |
| LMArena Hard Prompts | – | 1503 |
| LMArena Coding | – | 1527 |
| LMArena WebDev | – | 1540 |
| LMArena Vision | – | 1289 |
| AA Intelligence Index | 53.2 | 42.0 |
| IFBench | – | 62.2% |
| AA-LCR | 83.0% | 77.7% |
| AA-Omniscience | 42.4 | 28.8 |
| Terminal-Bench Hard | – | 58.3% |
| GPQA Diamond (AA) | 93.4% | 92.0% |
| Humanity's Last Exam (AA) | 58.7% | 48.7% |
| SciCode (AA) | 60.9% | 54.4% |
| τ²-Bench Telecom (AA) | – | 94.4% |
| LiveCodeBench | – | 87.8% |
| MMLU-Pro | – | 89.6% |
| LegalBench | – | 83.6% |
| CorpFin | – | 66.7% |
| TaxEval | – | 75.6% |
| Terminal-Bench 2.1 (Vals) | – | 71.9% |
| SWE-bench (Vals) | – | 88.6% |
| GPQA Diamond (Vals) | – | 92.4% |
| Vals Index | – | 60.9 |
| HiL-Bench | – | 35.3% |
| EQ-Bench 4 | – | 1281 |
| ARC-AGI-1 | 96.5% | – |
| ARC-AGI-2 | 90.0% | – |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.