Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, and described them as “the same model, but with different levels of safeguards.” Fable 5.1 is generally available on the Claude API as claude-fable-5-1 and on AWS, Google Cloud and Microsoft Azure from day one. Mythos 5.1, with some safeguards lifted, is reserved for trusted-access programs. The pattern is the one Anthropic set with Fable 5 and Mythos 5 in June, before a US government order suspended both and they came back in July.
The savings come from the cache
Per-token prices did not change: $10 per million input tokens and $50 per million output, as with Fable 5. What changed is the cache read, now $0.25 per million tokens, 75% less than before. Anthropic estimates that this cuts typical workloads by about 25% and highly agentic ones by up to about 45%, measured at default effort over four weeks of real usage in August. Those are its own estimates, and the saving depends on how much of a workload hits the cache. Anthropic also says that at low or medium effort Fable 5.1 matches or beats Fable 5 at lower cost.
Who can use Mythos 5.1
At launch, Mythos 5.1 access ran through the Life Sciences Verification Program, which had enrolled its first participants. For cybersecurity, Anthropic’s Cyber Verification Program still covered only Opus and Sonnet models, with Mythos-class access promised “in the near future.” Enterprise customers can get Mythos 5.1’s findings without direct access through Claude Security, a public beta that runs codebase scans on it.
Safeguards, retention and privacy
Fable 5.1 may now identify software vulnerabilities but not write exploits for them. Flagged requests fall back to another model, Opus 4.8 for cyber and Opus 5 for biology, and rerouted requests are not billed at Fable prices. Anthropic says its newest cyber safeguards produce 60% fewer false positives and its biology safeguards fire 85% less often on benign basic biology and medical requests.
Using a Fable-class model still requires 30-day data retention by default. The alternative, Enterprise Frontier Safeguards, keeps safety-monitoring data in the customer’s own cloud under the customer’s keys, but it rolls out in phases from the fall and was not available on September 1. Until then, eligible customers get zero data retention on Fable 5 and 5.1.
One change affects developers directly: to protect against distillation, API accounts created from September 1 can no longer edit Claude’s earlier context while keeping the thinking blocks of that transcript.
Marked output
Anthropic signed the EU code of practice on transparency of AI-generated content in July, which led it to watermark text from models released after August 2, 2026. Fable 5.1, released after that date, carries it from launch. The watermark adds no characters or cost and is not traceable to a user, but weakens on factual passages, light edits and code.
Whose numbers, and what the system card warns
The benchmark gains in the announcement are Anthropic’s own runs with production safeguards on, a setting that can lower scores where a safeguard intervenes. The independent GDPval-AA leaderboard from Artificial Analysis places Fable 5.1 first in knowledge work, ahead of Claude Opus 5. The system card adds its own warning: the model can still sometimes bypass approvals and auto-mode classifiers, and it is among the most capable Anthropic has tested at controlling what appears in its extended thinking.
Related reading
- Claude Fable 5.1 and Mythos 5.1 system card — Anthropic
- Claude Opus 5 keeps the Opus price — IA al Día
- Fable 5 is back with new safeguards — IA al Día