Claude Fable 5.1 and Mythos 5.1
Claude Fable 5.1 lifts Terminal-Bench 4.0 from 42.0% to 55.8% and cuts cache-read price 75%; Mythos 5.1 is the same model with fewer safeguards.
Same-tier upgrade of Fable 5, with Terminal-Bench-Science rising from 24.7% to 52.6% and CursorBench 3.2.0 from 70.5% to 73.4%. Safeguards were retuned, with cyber false positives down 60% (it may now find but not exploit vulnerabilities), a biology access program with the US government, and anti-distillation limits on editing prior thinking. Typical workloads about 25% cheaper.
- Date
- Tuesday, 1 September 2026
- Lab
- Anthropic
- Kind
- model
- Access
- closed API
- Price
- $10/$50 per M tokens in/out unchanged; cache reads $0.25 per M (was $1.00) (2026-09-01)
Figures
| Measure | Value | Measured by |
|---|---|---|
| Terminal-Bench 4.0 | 55.8% vs 42.0% Fable 5, 52.3% Opus 5, 37.3% GPT-5.6 Sol; Mythos 5.1 60.9% | company |
| Terminal-Bench-Science 0.1 | 52.6% vs 24.7% Fable 5, 29.0% Opus 5 | company |
| CursorBench 3.2.0 | 73.4% vs 70.5% Fable 5, 70.0% Opus 5 | company |
| Humanity's Last Exam (no tools) | 60.9% vs 57.8% Fable 5; 65.0% with tools | company |
| OSWorld 2.0 (strict) | 41.7% vs 36.1% Fable 5, 39.6% Opus 5 | company |
Fable 5.1 is generally available; Mythos 5.1 is limited to vetted US organizations through the Cyber Verification and Life Sciences Verification Programs. The comparison table also lists OpenAI's GPT-5.6 Sol. Cognition's quote says it would move Devin's Opus 5 traffic to Fable 5.1 on launch day. Enterprise Frontier Safeguards (zero-retention-equivalent monitoring) were announced alongside; see anthropic-enterprise-frontier-safeguards. Benchmarks are Anthropic runs with production safeguards on, which Anthropic says likely lowers scores.
Sources
- www.anthropic.com/claude-fable-and-mythos-5-1
- platform.claude.com/docs/en/models/overview
- platform.claude.com/docs/en/release-notes/overview
This record was checked against its sources on 6 October 2026. How we check