Claude Fable 5.1
Anthropic·
Overview
Successor to Claude Fable 5 in the Mythos-class tier, at the same $10/M input and $50/M output pricing but with cache reads cut to $0.25/M (2.5% of the input price instead of 10%). Launched alongside Claude Mythos 5.1, the same underlying model with safeguards lifted for Project Glasswing participants (tracked separately). Safety classifiers remain in place; refused requests can fall back server-side to Claude Opus 4.8 or Claude Opus 5. Adaptive thinking is always on, forced tool use is no longer supported, and all text output carries Anthropic's statistical watermark. Vendor-reported benchmarks: Terminal-Bench 4.0 55.8 (Fable 5: 42.0, Opus 5: 52.3), Terminal-Bench-Science 0.1 52.6, HLE 60.9 without tools (65.0 with tools), OSWorld 2.0 41.7 strict, CursorBench 3.2.0 73.4, AutomationBench 31.4.
Capabilities and innovations
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 93.43% | independent GPQA Diamond 93,43% (±1,86), measured by Vals AI at max effort (accessed 2026-09-01). Ties Claude Opus 5 on the same board. |
| Humanity’s Last Exam (no tools) | 60.9% | vendor HLE 60,9% without tools, 65,0% with tools — Anthropic launch announcement as reported in launch coverage (Decrypt, MarkTechPost, 2026-09-01). No-tools figure recorded for cross-model comparability. Artificial Analysis measures 59,1% at max effort with default fallback (58,7% xhigh, 55,9% high). |
| MMLU-Pro | 92.38% | independent MMLU Pro 92,38% (±0,27), rank 1 on the Vals AI MMLU-Pro leaderboard, max effort (accessed 2026-09-01). Not reported in Anthropic's launch material. |
| Terminal-Bench 2.x | 85.02% | independent Terminal-Bench 2.1 85,02% (±0,99), Vals AI at high effort, rank 2 behind GPT-5.6 Sol (85,77%), updated 2026-08-31. Anthropic itself reports Terminal-Bench 4.0 (55,8%), a different generation on a different scale — not comparable with 2.x values. |
| MMMU-Pro | 90.64% | independent MMMU Pro 90,64%, rank 1 on the Vals AI board (accessed 2026-09-01). |
- SWE-bench Verified: not reported by the vendor — Anthropic reports no SWE-bench Verified figure for this release.
- SWE-bench Pro: independent evaluation pending — No SWE-bench Pro figure found in launch coverage (Anthropic's headline coding number for this release is Terminal-Bench 4.0). Check the launch table and Scale's SEAL leaderboard once an independent run exists.
- MMLU: not reported by the vendor
- Terminal-Bench 3.0: independent evaluation pending — Terminal-Bench 3.0 public snapshot (2026-08-28) predates the release; no 3.0 value for Fable 5.1 yet. Anthropic reports Terminal-Bench 4.0 (55.8), a different generation that must not be entered in the 2.x or 3.0 fields.
API pricing
$10 per million input tokens, $50 per million output tokens (USD, cheapest OpenRouter endpoint, checked 2026-09-01)
Links
More from Anthropic
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.