Anthropic opens access to Claude Fable 5 while keeping Mythos 5 behind trusted controls

The same underlying model will be split between general API access and restricted use for selected cybersecurity and life sciences partners.

Anthropic has launched Claude Fable 5 for general use while keeping Claude Mythos 5 restricted to selected trusted access partners

Anthropic has launched Claude Fable 5 for general use while keeping Claude Mythos 5, the same underlying model with some safeguards lifted, restricted to selected cybersecurity, infrastructure, and future life sciences partners.

The launch gives developers access to Anthropic’s most capable generally available Claude model through the Claude API as claude-fable-5. Anthropic says Fable 5 outperforms its previous generally available models across software engineering, knowledge work, vision, scientific research, memory, and long-context tasks.

The release also introduces a more controlled access model for higher-risk capabilities. Fable 5 uses safeguards that redirect some requests in cybersecurity, biology and chemistry, and model distillation to Claude Opus 4.8 instead of allowing the main model to respond.

Mythos 5 is being made available first to existing Claude Mythos Preview users, including partners in Project Glasswing, Anthropic’s program for cyberdefenders and critical software infrastructure providers. Anthropic says it plans to expand Mythos 5 access through broader trusted access programs for cybersecurity organizations and selected biology researchers.

Both models are priced at $10 per million input tokens and $50 per million output tokens. Anthropic says that is less than half the price of Claude Mythos Preview.

General release with fallback safeguards

The key difference between Fable 5 and Mythos 5 is not the underlying model, but who can use which capabilities and under what conditions.

Fable 5 has been launched with classifiers that detect requests Anthropic considers higher-risk. When those classifiers are triggered, the request is handled by Claude Opus 4.8. Users are told when this happens.

Anthropic says the safeguards trigger in less than five percent of sessions on average and that more than 95 percent of Fable 5 sessions involve no fallback. The company also says the safeguards have been tuned conservatively, so some harmless requests may be caught.

The areas covered by the classifiers include offensive cybersecurity tasks, biology and chemistry requests, and attempts to distill Claude’s capabilities into other models. Anthropic says the safeguards are intended to limit misuse while allowing general access to a Mythos-class model.

The company says its external bug bounty produced no universal jailbreaks in more than 1,000 hours of testing. Anthropic also says external red-teaming organizations did not find universal jailbreaks on long-form agentic tasks, although the UK AI Security Institute made progress toward one during a short initial testing window.

Software engineering, vision, and research claims

Anthropic is positioning Fable 5 around longer tasks where models need to sustain context, use memory, and complete multi-stage work.

In early testing cited by Anthropic, Stripe reported that Fable 5 completed a codebase-wide migration in a 50-million-line Ruby codebase in one day. Stripe said the work would otherwise have taken a team more than two months by hand.

Anthropic also says Fable 5 scored highest among frontier models on Cognition’s FrontierCode evaluation at medium effort. The evaluation tests whether models can complete difficult coding tasks while meeting production codebase standards.

For knowledge work, Anthropic says Fable 5 achieved the highest score of any model on Hebbia’s Finance Benchmark for senior-level reasoning. IMC also reported that Fable 5 performed strongly in trading-analysis evaluations covering factual lookup, conceptual reasoning, root-cause analysis, and expected-value analysis.

Anthropic says the model has improved vision capability, including extracting numbers from scientific figures and rebuilding a web app’s source code from screenshots. The announcement also says Fable 5 completed Pokémon FireRed using raw game screenshots and a minimal vision-only harness.

Restricted access for Mythos 5

Mythos 5 is being held back from general release because Anthropic says Mythos-class models present significant risks in areas including cybersecurity and research biology.

Project Glasswing partners can upgrade from Claude Mythos Preview to Mythos 5 from launch. Anthropic says Mythos 5 is comparable to, or stronger than, Mythos Preview in most cases and costs substantially less.

Anthropic plans to expand access to cybersecurity organizations through a trusted access program developed in consultation with the US government. A separate biology program will give selected life sciences researchers access to Fable 5 with biology and chemistry safeguards removed, while cyber safeguards remain in place.

Anthropic says Mythos 5 accelerated aspects of internal drug design work by around ten times and matched or beat skilled human operators in one protein design example. The company also says its scientists preferred Mythos 5’s molecular biology hypotheses around 80 percent of the time in blinded comparisons against Opus-class models.

In genomics, Anthropic says Mythos 5 completed more than a week of mostly autonomous research, using single-cell data from millions of cells across 138 animal species to train a model for identifying similar cell roles across organisms. Anthropic says it intends to publish the results in the coming months.

Anthropic is also changing data retention for Fable 5, Mythos 5, and future models at similar or higher capability levels. Business customer traffic on Mythos-class models will be retained for 30 days for safety purposes, but Anthropic says the data will not be used to train new Claude models or for non-safety-related purposes.

Fable 5 is available from 9 June through the Claude API and consumption-based Enterprise plans. Pro, Max, Team, and seat-based Enterprise plans include Fable 5 at no extra cost until 22 June, with usage credits required from 23 June unless Anthropic extends the included access window.