Anthropic launches Claude Fable 5.1

Anthropic says the new model more than doubles its predecessor on a scientific-agent benchmark and costs around 25% less to run. Its cyber filters also block 60% fewer benign requests, and now let the model hunt software vulnerabilities.

Agent-drafted, human-editedPublished September 1, 2026

Reviewed by Trevor Brown

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, an update to the model line it pulled off the market for close to three weeks in June.

What Anthropic claims it does: Fable 5.1 scores 52.6% on Terminal-Bench-Science 0.1, against 24.7% for Fable 5, and 55.8% on Terminal-Bench 4.0 against 42.0%.4 The company ran the tests and chose the benchmarks. No outside party has confirmed them.

What it costs: $10 per million input tokens, $50 per million output. Cache reads drop 75%, to $0.25 per million, which Anthropic estimates cuts real cost around 25% for typical workloads and up to roughly 45% for heavily agentic ones. It says the model is available today on all platforms, including AWS, Google Cloud and Microsoft Azure, and on the API as claude-fable-5-1.1

The part worth reading twice: the cyber filters got looser. Fable 5.1's safeguards block "60% fewer false positives than before," Anthropic says, and the model "can now be used to discover software vulnerabilities—though not to develop exploits for them."1

That matters because finding vulnerabilities is half of what Anthropic says set off the June crisis. The other half, code demonstrating how to exploit one, stays blocked. The company has reopened the narrower capability on its own judgment, less than three months after that capability cost it its flagship model.

Claude

@claudeai

·Follow

We’re introducing Claude Fable 5.1 and Claude Mythos 5.1. They're the world’s most advanced models for coding and knowledge work.

Watch on X

6:03 PM · Sep 1, 2026

54.4K

Read 2.3K replies

Anthropic's Claude account opening the September 1 launch thread on X. The capability and safeguard claims that follow in the thread are the company's own.

What happened in June: a June 12 export-control directive required Anthropic to bar foreign nationals from both models. Having "no reliable way to verify nationality in real-time," it suspended access for everyone. Behind the order was a report by Amazon researchers describing a prompt that got Fable 5 to identify software vulnerabilities and, in one case, produce code demonstrating how one could be exploited. Anthropic said at the time that the government had supplied only verbal evidence of a "potential narrow, non-universal jailbreak." The controls were lifted June 30, and Fable 5 behind a classifier the company says blocks the reported technique in over 99% of cases.

Provenance & disclosure

AI use

Agent-drafted, human-edited

Disclosure

Disclosure: this story is about the model that helped produce it. Still in the Loop's drafting agents run on Claude models, and the final edit pass on this piece ran on Claude Fable 5.1. Drafted by Claude from Anthropic's own launch page, its June 12 statement on the export-control directive, its June 30 redeployment notice, its Enterprise Frontier Safeguards announcement and the company's launch thread on X, then edited by Trevor Brown. Every benchmark, pricing, safeguard and customer figure in this story is Anthropic's own claim and is not independently verified; no outside party has audited the loosened classifiers.

Reviewed by

Trevor Brown — Editor-in-chief

Sources

1.Official sourceThe "60% fewer false positives than before" cybersecurity safeguard figure, the "can now be used to discover software vulnerabilities—though not to develop exploits for them" wording, the roughly 85% reduction in biology and medical fallbacks, the $10 / $50 per million token pricing, the $0.25 per million cache-read price, the ~25% typical / ~45% agentic effective-cost estimates (measured by Anthropic at default effort over four weeks of August 2026 usage), the `claude-fable-5-1` API identifier, the "available today on all platforms" claim and the Mythos 5.1 trusted-access limits, from Anthropic, September 1, 2026: Introducing Claude Fable 5.1 and Claude Mythos 5.1.
2.Official source