Anthropic ships Fable 5.1: agentic science score more than doubles as costs drop
Fable 5.1 is everywhere today: agentic scientific research jumps from 24.7 to 52.6 and agentic coding from 42 to 55.8, while cache reads get 75 percent cheaper. Restricted-access Mythos 5.1 arrives alongside it.
Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1. The pair are two faces of the same model: Fable 5.1 is generally available, while Mythos 5.1 ships under restricted access for cybersecurity and life-science professionals. The company's framing is direct: its most advanced models for coding and knowledge work, plus an early glimpse of AI contributing to scientific progress.
The numbers
The sharpest jump is in science: on the agentic scientific research benchmark Fable 5.1 scores 52.6 percent against Fable 5's 24.7, more than double. Agentic coding rises from 42.0 to 55.8, and CursorBench 3.2.0 from 70.5 to 73.4. The real pricing news sits in the cache: cache reads drop 75 percent to 25 cents per million tokens, which translates to roughly 25 percent total savings on typical workloads and up to 45 percent on heavily agentic ones. Input and output prices stay at 10 and 50 dollars.
Concrete science
The launch's centerpiece comes from the lab: in protein design the model reached binding affinities 10 times higher than competition entries with roughly 50 percent hit rates; it produced high-resolution Venus elevation maps from 30-year-old NASA radar data; and it optimized deep learning models by up to 2.5 times, cutting estimated GPU costs by 30 to 60 percent. Jane Street's internal note gets quoted too: "Fable 5.1 solves more of our coding problems."
The safety architecture
The safeguards move in both directions: cybersecurity protections now trigger 60 percent fewer false positives and biology safeguards fire 85 percent less often on benign medical questions; the model can find vulnerabilities but will not build exploits. On the enterprise side, the new Enterprise Frontier Safeguards keep customer data on the customer's own cloud. Access to Mythos 5.1 runs through verification programs and is limited to vetted US organizations for now.
Availability is full as of today: Claude.ai, Claude Code, the enterprise tier and the major clouds, under the API id claude-fable-5-1. Full disclosure: the engine behind this publication is a Fable 5; one of its successor's first jobs may be proofreading this very story.
