Anthropic Releases Claude Fable 5.1 and Claude Mythos 5.1: 52.6% on Terminal-Bench-Science and 75% Cheaper …
What changed
Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, iterations of the same foundational AI model but deployed under different access and safety conditions. Fable 5.1 is broadly available via the Claude API and through major cloud platforms like AWS, Google Cloud, and Microsoft Foundry. Mythos 5.1 remains locked behind strict vetting under Project Glasswing, reserved for select organizations. Fable 5.1 boosts the Terminal-Bench-Science 0.1 benchmark score from 24.7% to 52.6%, doubles the usability with a 1 million token context window, and slashes cache read costs by 75%, lowering charges to $0.25 per million tokens without raising base prices.
Why builders should care
The improved benchmark reflects more capable reasoning and knowledge handling, which supports more complex AI tasks and workflows. The expanded context window means applications can feed the model far larger inputs or maintain longer conversations without additional prompting overhead. The sharp reduction in cache read costs makes handling high-volume token use cases much cheaper, especially for builders relying on large context interactions or persistent chat histories. These cost efficiencies are critical for scaling AI services without ballooning infrastructure bills.
The practical takeaway
Builders get a more powerful and cheaper AI tool to integrate. The 1 million token context window enables use cases such as extended document analysis, multi-turn dialog spanning hours or days, and complex multi-stage workflows that previously strained token limits. The 75% cost cut in cache reads means long sessions or frequent repeated queries become financially viable. For those using Claude Fable, this update directly lowers operational costs while boosting model performance, making it a smarter pick for scaling AI-heavy products and prototypes.
What to watch next
It is important to track how widely Mythos 5.1 opens beyond Project Glasswing. That version’s controlled access suggests added safety or policy features worth watching for potential regulatory or enterprise uses. Watching pricing shifts post-launch will reveal if Anthropic sustains these cost-per-token improvements industry-wide. Builders should also monitor developer feedback on large context window use in production, as this could reshape best practices for prompt engineering and session management.
AI Quick Briefs Editorial Desk