Anthropic Releases Claude Fable 5.1 and Claude Mythos 5.1: 52.6% on Terminal-Bench-Science and 75% Cheaper Cache Reads
MarkTechPost Asif Razzaq ● Covered by 2 sources
Anthropic just shipped Claude Fable 5.1 and Mythos 5.1. Same model, different guardrails — and cache reads got 75% cheaper.
Based on reporting by MarkTechPost, Asif Razzaq — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, three months after the Fable 5 line arrived in June 2026. Under the hood, they’re the same model. The split is in the safeguards: Fable 5.1 is the public release, while Mythos 5.1 stays limited to vetted organizations inside Project Glasswing.
The headline number is a big one. On Terminal-Bench-Science 0.1, Anthropic says Fable 5.1 reaches 52.6%, up from 24.7% for Fable 5 and 29.0% for Opus 5. The company also puts GPT-5.6 Sol at 22.4% on the same benchmark. Anthropic adds a standard error of 3.5 to 4.5 points per model, which is a useful reminder that the exact ordering matters less than the size of the jump.
There’s a second story here, and it’s the one teams will probably feel first: price. Cache reads drop from $1.00 to $0.25 per million tokens, a 75% cut. Anthropic says that works out to roughly 25% lower cost on typical workloads and up to about 45% on agent-heavy ones, while base input and output pricing stays at $10 and $50 per million tokens. Batch processing is set at $5 and $25 per million tokens.
Availability is split the same way as the safeguards. Fable 5.1 is generally available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry. Mythos 5.1 is not broadly open; it’s restricted to vetted US organizations inside Project Glasswing. Both models share a 1M token context window, a 128K maximum output, and adaptive thinking that is always on.
Anthropic is also warning developers about a few breaking changes. Forced tool use is gone, thinking blocks are now tied to the model, and edits to earlier turns can invalidate those blocks. The company says the checks are enforced for accounts created on or after August 31, 2026, and points teams toward turn-scoped system messages and server-side context editing. On the safety side, Anthropic says cyber safeguards now allow vulnerability discovery but not exploit development, while biology safeguards trigger far less often on benign requests.
My take — AI-written commentary, not fact-checked reporting
This is the kind of release that quietly matters more than the flashy benchmark chart. A cheaper cache and stricter conversation rules are exactly the sort of unsexy changes that separate a demo toy from a system people can actually build on. Also, calling the safeguard gap “an unusually honest disclosure” feels like the bar for AI transparency is still painfully low.
Read more about this at: MarkTechPost