Anthropic Ships Claude Fable 5, Its Most Capable Public Model

Fable 5 is the most capable model Anthropic has ever released publicly

Anthropic launched Claude Fable 5 on June 9, 2026, making the first Mythos-class model available to all users. Mythos-class sits above Opus in Anthropic's capability hierarchy. Until now, Mythos-level capability was restricted to a narrow group of US government-affiliated cyber defenders through Project Glasswing, which began in April 2026.

Fable 5 leads on nearly every major benchmark Anthropic tested. In software engineering, Stripe reported that the model compressed months of engineering into days, completing a codebase-wide migration across a 50-million-line Ruby repository in a single day. On Cognition's FrontierCode evaluation, which tests model performance against production-grade coding standards, Fable 5 scores highest among all frontier models. In knowledge work, it posted the top score on Hebbia's Finance Benchmark for senior-level analytical reasoning. The model also introduces persistent memory improvements, using file-based notes to improve its own outputs over the course of long-running autonomous tasks.

Anthropic built new classifiers to make Mythos capability available without unrestricted access

The release hinges on a new safeguard architecture. Fable 5 uses safety classifiers that detect potential misuse, including jailbreak attempts, and automatically reroute those queries to Claude Opus 4.8 rather than refusing outright. The classifiers cover three areas: cybersecurity exploitation, biology and chemistry dual-use research, and requests that pattern-match as attempts to distill Claude's capabilities into competing models.

Anthropic deliberately tuned these classifiers conservatively, meaning they will occasionally catch harmless requests to ensure they catch enough harmful ones. An external bug bounty produced no universal jailbreaks in over 1,000 hours of testing. The UK AI Safety Institute made limited progress toward a partial jailbreak within a brief testing window, a result Anthropic disclosed publicly in the system card.

A second model, Claude Mythos 5, launches alongside Fable 5 with cyber safeguards removed for pre-approved partners only. Anthropic describes Mythos 5 as having the strongest cybersecurity capabilities of any model currently available. Access is restricted to Glasswing partners initially, with a broader trusted access program for biomedical researchers forthcoming.

Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, less than half the cost of the earlier Claude Mythos Preview. Subscription users on Pro, Max, Team, and Enterprise plans have free access through June 22. On June 23, Anthropic removes Fable 5 from those plans and requires usage credits, with the stated intent to restore it as a standard subscription feature once capacity allows.

For the first time, a frontier AI lab has released two versions of the same underlying model with different safety profiles, treating restrictions as a configurable runtime layer rather than a fixed property of the model itself. Every lab building at the frontier will now face pressure to explain why they have not applied the same distinction.