Claude Fable 5.1 Lands With 75% Cheaper Cache Reads and a Doubled Science Score

Anthropic released Claude Fable 5.1 and its restricted twin Mythos 5.1, keeping the headline token price at $10 per million input and $50 per million output while cutting cache reads by 75% to $0.25 per million, saving roughly 25% on typical workloads and up to 45% on context-heavy coding. Fable 5.1 scores 52.6% on Terminal-Bench-Science 0.1, up from 24.7% for Fable 5, and 55.8% on Terminal-Bench 4.0, while Mythos 5.1 reaches 60.9%. Anthropic also reported that Mythos 5.1 wrote custom GPU kernels speeding up seven open-source biology models by up to 2.5x and produced a new Venus elevation model from NASA Magellan data.

Claude Fable 5.1 keeps the same headline token price. The cache is what got cheap. Anthropic shipped Claude Fable 5.1 today, alongside a restricted twin called Mythos 5.1, and the most interesting number is not on the benchmark table. It is on the invoice. Fable 5.1 keeps the same headline pricing as Fable 5 — $10 per million input tokens, $50 per million output — but cache reads, where the model re-reads context it has already processed, fall 75% to $0.25 per million. On agentic work the cache is most of what you pay for, so Anthropic puts the saving at roughly 25% for typical workloads and up to about 45% for context-heavy, tool-heavy coding. The capability jump is real as well, and lopsided. On Terminal-Bench-Science 0.1, an agentic scientific research benchmark, Fable 5.1 scores 52.6% against 24.7% for Fable 5 and 29.0% for Opus 5. Terminal-Bench 4.0 goes from 42.0% to 55.8%, with the less-restricted Mythos 5.1 reaching 60.9%. Elsewhere the gains are far more modest: CursorBench 3.2.0 moves from 70.5% to 73.4%, and Humanity’s Last Exam from 57.8% to 60.9% without tools. Worth knowing before you compare notes with anyone: Fable 5.1 defaults to high effort in Claude Code but medium in Claude Cowork and on claude.ai, and the effort tier moves both the score and the cost. The part that touches hardware Tucked into the research section is the result most relevant to this audience. Working from publicly available source code alone, Mythos 5.1 wrote custom GPU kernels and cached intermediate results for seven open-source deep learning models used in computational biology, speeding them up by as much as 2.5x while producing identical outputs. Anthropic estimates that cuts GPU cost on genome-wide analyses by 30 to 60%, reckons the same work would normally occupy a team of performance engineers for weeks, and says it intends to open-source the optimisations. The other showpiece is a map. Fable 5.1 trained a neural network on radar data from NASA’s Magellan mission, which ended in 1994, and produced a new elevation model covering about a third of Venus at 2 to 3 km resolution rather than 10 to 20 km, with heights up to 25% more accurate. Anthropic is releasing it under a Creative Commons licence ahead of NASA’s VERITAS and ESA’s EnVision missions. Mythos is still fenced off Fable 5.1 and Mythos 5.1 are the same underlying model with different safeguards, which is the arrangement that got awkward for Anthropic earlier this year. Mythos 5.1 stays limited to vetted cyberdefenders and life scientists, and for now to US organisations only, with the company saying it is coordinating with the US government to widen access. Fable 5.1 itself has loosened: Anthropic says its cybersecurity safeguards now block 60% fewer false positives, partly because Fable 5.1 is permitted to find software vulnerabilities — though still not to build exploits for them. Two more things worth flagging. Enterprise Frontier Safeguards, the company’s answer to customers who want zero data retention, keeps data in cloud infrastructure the customer controls rather than Anthropic’s; it arrives in phases from later this autumn, with plain zero-retention offered to eligible customers until then. And under the EU AI Act code of practice Anthropic signed in July, models released after 2 August 2026 carry an invisible watermark, with a detection API now in private preview for regulators, researchers, media and fact-checkers. None of the benchmark figures are independently verified yet, and they come with Anthropic’s own caveats: several were run with production safeguards enabled, which zeroed some tasks outright and, by the company’s own admission, drags its scores down in ways that make cross-vendor comparison messy. Fable 5.1 is available today on the Claude API as claude-fable-5-1 , plus AWS, Google Cloud and Microsoft Azure. Source: Anthropic https://www.anthropic.com/claude-fable-and-mythos-5-1 .