Fable 5.1 and Mythos 5.1 released
On 1 September 2026 Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, which it calls the same model with different safeguards. Fable 5.1 is generally available through the Claude API and the three large clouds; Mythos 5.1 is open only to vetted cyberdefenders and life scientists through trusted access programs. The price stays at $10 and $50 per million input and output tokens, with cache reads cut to $0.25.
Why it matters
Two releases of one model: what separates the public version from the restricted one is the safeguards, not the weights. Alongside the model Anthropic changed the terms around it: a cheaper cache, fewer false cyber blocks and a promised privacy tier for enterprises. The company itself presents these as a response to customer feedback.
What the announcement says. Fable 5.1 and Mythos 5.1 are 'the same model, but with different levels of safeguards'. Fable 5.1 is available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry (id claude-fable-5-1), with a 1M-token context and up to 128K tokens of output. Mythos 5.1 is available only through the Cyber Verification Program and the Life Sciences Verification Program, for now to a set of US organisations; the API release notes say 'for Project Glasswing participants'. Price. Token prices are unchanged ($10 per million input tokens, $50 output); cache reads cost $0.25, 75% less. By Anthropic's measurement over four weeks of August 2026 that comes to about 25% lower cost for a typical workload and up to about 45% for the most agentic work. Safeguards. In Claude Code the cyber safeguards, the company says, intervene about 60% less often per session; finding software vulnerabilities with Fable 5.1 is now allowed, while exploit generation, penetration testing and binary-based scanning are still redirected to Opus models. The biology safeguards fire 85% less often on benign elementary biology and medical questions; life-sciences research is still directed to Opus. New API accounts can no longer edit Claude's prior thinking in a conversation, which closes a known distillation method. Enterprise Frontier Safeguards (customer data stays in the customer's own cloud) was announced, to roll out in phases from the fall. The company's scientific claims. Using open protein-design tools, Mythos 5.1 proposed binders for 12 targets; its hit rate was nearly 50% (typically 10-15%), and for three targets the affinities were 10 times higher than the best designs from Adaptyv Bio's competitions; the designs were sent to two external organisations for lab validation. Fable 5.1 trained a neural network to make an elevation map of a third of Venus from Magellan radar data, at 2-3 km resolution instead of 10-20. What the record does not claim. All figures (for example Terminal-Bench-Science 0.1: 52.6% against Fable 5's 24.7%) are Anthropic's own measurements, made with production safeguards on, which the company says probably lower the score. No independent evaluation was read. The announcement page carries no date: 1 September comes from the API release notes and the first Wayback capture; the Opus 5.5 page mentions 31 August only as the cut-off for the account rule.