Anthropic
Claude Fable 5 benchmark results
12 results on 7 benchmarks from 4 sources.
Also listed as: claude-fable-5-high · claude-fable-5_max · Claude Fable 5 (max) · claude-fable-5_low · Claude Fable 5 (low) · claude-fable-5_high · Claude Fable 5 (high) · Claude Fable 5 (Max) · Claude Fable 5 (XHigh)
All benchmark results
| Benchmarks | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|
| LMArena Text | 1504 | 4 | Overall, style control on; 40026 votes; 95% CI 1499–1508“claude-fable-5-high” | Official leaderboard | Oct 8, 2026 |
| LMArena WebDev | 1625 | 21 | Overall; 14657 votes; 95% CI 1618–1631“claude-fable-5-high” | Official leaderboard | Oct 8, 2026 |
| LMArena Vision | 1309 | 1 | Overall, style control on; 13709 votes; 95% CI 1301–1316“claude-fable-5-high” | Official leaderboard | Oct 2, 2026 |
| GPQA Diamond | 85.9% | 77 | Epoch AI run, accuracy; effort: max | Independent run | Aug 6, 2026 |
| GPQA Diamond | 83.3% | 96 | Epoch AI run, accuracy; effort: high | Independent run | Aug 6, 2026 |
| GPQA Diamond | 78.8% | 121 | Epoch AI run, accuracy; effort: low | Independent run | Aug 6, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 100.0% | 1 | Epoch AI run, OTIS Mock AIME 2024–2025; effort: high | Independent run | Aug 6, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 99.7% | 13 | Epoch AI run, OTIS Mock AIME 2024–2025; effort: max | Independent run | Jun 10, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 97.8% | 31 | Epoch AI run, OTIS Mock AIME 2024–2025; effort: low | Independent run | Aug 6, 2026 |
| ARC-AGI-1 | 98.5% | 1 | Semi-private set; $2.11/task; CoT“Claude Fable 5 (Max)” | Official leaderboard | Oct 22, 2026 |
| ARC-AGI-1 | 98.5% | 2 | Semi-private set; $1.02/task; CoT“Claude Fable 5 (XHigh)” | Official leaderboard | Oct 22, 2026 |
| τ²-bench | 39.7% | 8 | banking_knowledge domain, pass^1; effort: max; submitted by Sierra; v1.0.1 | Official leaderboard | Jul 23, 2026 |
FAQ
Where do these Claude Fable 5 scores come from?
From Epoch AI Benchmarking Hub, ARC Prize leaderboard, LMArena leaderboard dataset, and τ²-bench leaderboard submissions. Each row links to the original leaderboard, dataset or report, and vendor-reported numbers are labelled.