Anthropic released Claude Fable 5 and Claude Mythos 5 on 9 June — its next generation of intelligence for the hardest knowledge work and coding problems. Fable 5 is the public release, available immediately via the Claude API, while Mythos 5 retains lifted safeguards for approved researchers and cybersecurity professionals through Project Glasswing.
The software engineering capabilities are the headline. Fable 5 compressed what Anthropic described as 'months of engineering into days', completing a 50-million-line Ruby codebase migration in a single day — a task that would have required two months of team effort. It scored highest among all frontier models on Cognition's FrontierCode evaluation and drew praise from senior leaders at Cursor, GitHub, Replit, and Perplexity as a 'clear step forward' with improved autonomy on long-horizon coding tasks.
The vision capabilities are equally striking. Fable 5 achieved state-of-the-art performance extracting precise numbers from scientific figures and rebuilding web application source code from screenshots alone. In a demonstration of sustained visual reasoning, the model completed Pokémon FireRed using only raw game screenshots without additional tools.
In the life sciences, Mythos 5 accelerated protein design processes by approximately tenfold and produced novel molecular biology hypotheses that researchers preferred 80 per cent of the time in blind comparisons. In genomics, it conducted autonomous research identifying cells performing similar roles across 138 animal species. For long-context work, Fable 5 showed a threefold performance improvement over Opus 4.8 when given persistent file-based memory.
Pricing is set at $10 per million input tokens and $50 per million output tokens — less than half the cost of Claude Mythos Preview. The safety architecture uses AI classifiers that detect potential misuse and fall back to Opus 4.8 rather than refusing outright, activating in fewer than 5 per cent of sessions. External bug bounties with over 1,000 hours of testing found no universal jailbreaks, and an independent partner confirmed Fable 5 complied with zero harmful requests across 30 public jailbreak techniques.
For context engineers, Fable 5 represents the most significant capability jump since Opus 4.5. The model's ability to autonomously handle multi-day engineering tasks, reason across millions of tokens of context, and operate with visual understanding of application state suggests that AI-assisted development is moving from pair programming toward genuine delegation of complex engineering work.
