Today’s edition of my newsletter just went out.
Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-06-09
Anthropic publicly releases Claude Fable 5 with classifier-based safety routing that falls back to Opus 4.8 for sensitive requests, while Cognition simultaneously launches FrontierCode, a benchmark testing whether AI-generated code meets real-world merge standards.
Appears in
Extraction
Topics: claude-fable-5ai-safetymodel-routingcoding-benchmarksfrontier-ai
Claims
- Fable 5 and Mythos 5 share one underlying model, with Fable adding classifier gates for the general public while Mythos lifts some restrictions for vetted partners.
- When classifier gates trigger on sensitive topics such as cyber, biology, chemistry, or model-copying, Fable 5 routes the request to Claude Opus 4.8 rather than issuing a direct refusal.
- Fable 5 supports a 1M-token context window and was used to complete a 50M-line Ruby migration in one day.
- API pricing for Fable 5 is $10 per million input tokens and $50 per million output tokens.
- Cognition's FrontierCode benchmark evaluates whether AI-generated code is good enough for a real maintainer to merge, not just whether it passes automated tests.
Key quotes
Fable 5's capabilities exceed those of any model we've ever made generally available.
the product is no longer just a model, but a routing machine that decides which level of intelligence a user is allowed to touch for each request.
We used to check if Claude is doing the work right… With Claude Fable 5, I instead check if Claude is doing the right work