Claude Fable 5 and Mythos 5, Anthropic’s new most capable models

Claude Fable 5 and Claude Mythos 5 share the same underlying model and belong to the Mythos category, Anthropic's highest tier, above the Opus class. The difference lies in the restrictions applied: Fable 5 incorporates security classifiers that limit certain responses in risk-sensitive areas, while Mythos 5 has those restrictions partially lifted and is reserved for controlled access groups. One of the most notable improvements is the ability to work autonomously over extended periods: the longer and more complex the task, the greater the advantage these models have over previous generations.

In autonomous coding, Fable 5 reaches 80.3% on SWE-Bench Pro, compared to 69.2% for Opus 4.8 and 58.6% for GPT-5.5. Stripe, during pre-launch testing, used the model to migrate a 50-million-line Ruby codebase in a single day — a task that would have taken a full team more than two months by hand. In vision capabilities, the model completed the video game Pokémon FireRed using only raw screenshots, with no additional tools, something previous models could not achieve even with assistance. In advanced financial reasoning tasks, it ranks first on Hebbia's benchmark for senior-level analysts.

In scientific research, Mythos 5 performed protein design for 14 therapeutic targets without human assistance, yielding viable candidates for 9 of them. The model also conducted autonomous genomics research over more than a week, training a custom model that outperformed one published in the journal Science while being one hundred times smaller.

Fable 5's safeguards work through independent classifiers that detect queries related to offensive cybersecurity, biology, chemistry, and capability distillation, redirecting those requests to Claude Opus 4.8 rather than blocking them outright. According to Anthropic, fewer than 5% of sessions trigger this mechanism. The classifiers underwent more than 1,000 hours of external penetration testing with no universal bypass found. In internal evaluations, the success rate on offensive cyber tasks drops to 5.4% with Fable 5, compared to 56.6% with Opus 4.8. On alignment, Mythos 5's metrics are similar to those of Opus 4.8 according to Anthropic's automated assessments. Mythos 5, with cyber restrictions lifted, is deployed through Project Glasswing in collaboration with the US government.

Both models are priced at $10 per million input tokens and $50 per million output tokens — less than half the price of Claude Mythos Preview. Fable 5 is available today for all users; Mythos 5 is limited to current Glasswing partners and, soon, to a trusted access program for biology researchers.

Key points

• Fable 5 and Mythos 5 are the same model with different safeguards
• They belong to the new Mythos category, above the Opus class
• They stand out for working autonomously on long and complex tasks
• Fable 5 leads in autonomous coding: 80.3% on SWE-Bench Pro vs. 69.2% for Opus 4.8
• Stripe migrated 50 million lines of code in one day using Fable 5
• Mythos 5 designed proteins for 9 of 14 therapeutic targets without human assistance
• Safeguards redirect sensitive queries to Opus 4.8; fewer than 5% of sessions are affected
• Mythos 5 operates through Project Glasswing with the US government
• Price: half that of Mythos Preview; Fable 5 available today for everyone