Anthropic Ships Claude Fable 5.1 and Mythos 5.1, With Cache Reads Cut 75% and the Strongest Tier Gated

On September 1, 2026 Anthropic released two models from its Claude 5.1 generation: Claude Fable 5.1, generally available, and Claude Mythos 5.1, reachable only by application. DoNews reported the pair as aimed at coding and knowledge work, with the Mythos line opened to vetted security and life-sciences organisations.

The release is as much about distribution as capability

• One generation, two access tiers: a public model and a gated one, the latter granted through a "trusted access" programme.
• Reporters place Mythos 5.1's eligibility in cybersecurity and the life sciences — consistent with the frontier-capability tiers Anthropic has described for itself.
• When this site covered OpenAI's GPT-6 Astra, OpenAI had benchmarked against Claude Fable 5.1; each lab now measures itself against the other's flagship.

Where the price cut lands

Outlets including DoNews, Sina Tech and AIhot agree: cache-read pricing falls by roughly 75%. That is the line item that matters most to applications that re-read the same long context — codebase retrieval, long-document question answering and agent loops — because it reduces the cost of repetition rather than the price of a first input.

On benchmarks, the reports this site checked name Terminal-Bench-Science and GDPval-AA but give no figures a third party could re-verify, so those scores are left as vendor claims and are not repeated here as fact.

Rumours about what comes next

Third-party trackers and community threads are already discussing a 5.2 and a stronger internal model. None of it is confirmed by Anthropic, so this article treats it as rumour rather than reporting.

Pricing and eligibility are as stated in DoNews's report and on Anthropic's own pages; on the cache-price change see also AIhot's summary.