Claude Fable 5.1 and Mythos 5.1 ship for coding and long-horizon agents

Anthropic launched Claude Fable 5.1 as a generally available model on September 1, 2026, and Claude Mythos 5.1 as a limited-access twin. The company says they are the same model with different safeguards: Fable 5.1 is the public path for coding, knowledge work, and long-running agent loops; Mythos 5.1 keeps more permissive cybersecurity and life-sciences settings for trusted-access programs. Amazon Web Services listed Fable 5.1 the same day on Amazon Bedrock and Claude Platform on AWS, and described Mythos 5.1 as available with limited access and its full cyber and bio capabilities retained.

For teams running Claude Code, Cowork, or custom tool loops, this is a Fable-class upgrade with cheaper cache reads. It also arrives as a Covered Model, with tighter thinking-transcript controls and cyber and biology redirects that can still interrupt a naive agent harness.

Cheaper cache reads

Anthropic estimates Fable 5.1 will cost about 25 percent less than Fable 5 on typical token-billed workloads because it is cutting the price of cache reads—inputs the model has already processed and stored. For highly agentic work, it puts the savings at up to about 45 percent. The Verge independently recapped those estimates. Anthropic’s launch post does not publish a new headline input/output rate.

That math is why Cognition is treating the release as a same-day swap. Co-founder and CPO Walden Yan said the company is moving Opus 5 traffic in Devin to Claude Fable 5.1 on launch day: it matched or edged out Fable 5 in testing at a lower cost per task, “and with the new cache read pricing a Fable-class model is finally economical for the workloads we’d kept on Opus, starting with code review.”

Claude Code, effort defaults, and benches

Fable 5.1 defaults to High effort in Claude Code, and to Medium in Claude Cowork and on Claude.ai. Anthropic staffer Cat Wu posted that the company’s own team has used the model on more ambitious projects that would previously have taken months, and told practitioners to try it in Claude Code, Claude Cowork, and Claude Tag.

On Anthropic’s Terminal-Bench 4.0 table, Fable 5.1 scores 55.8 percent and Mythos 5.1 scores 60.9 percent, versus 42.0 percent for Fable 5. Anthropic says the Fable–Mythos gap reflects tasks where earlier, less precise cyber safeguards intervened, and that it expects the difference to be much smaller with the safeguards shipping today. At Low or Medium effort, it says Fable 5.1 matches or beats Fable 5 at lower cost. Other scores in the same table include 52.6 percent on Terminal-Bench-Science 0.1, against 24.7 percent for Fable 5, and 73.4 percent on CursorBench 3.2.0, against 70.5 percent.

AWS describes the model as built for long-running work that spans hours and many applications, and says it is less likely to take shortcuts such as disabling a failing test. Early-access partners leaned on the same long-horizon point. Jane Street’s Craig Falls said that while prior models became hard to follow the longer they worked, Fable 5.1 “remains readable over long, multi-step tasks.” Shopify’s Ben Lafferty said it is more comfortable with long, unattended work than Fable 5: it keeps its own records, reprioritizes as things change, and picks up where it left off.

Covered Model, Mythos access, and harness gotchas

AWS says Anthropic designated Fable 5.1 a Covered Model, a category with additional data-retention, safety-review, and access policies wherever it is offered. Anthropic’s Enterprise Frontier Safeguards (EFS) are meant to give enterprise customers the privacy of a zero-data-retention policy by storing data in cloud infrastructure the customer controls. EFS rolls out in phases beginning later this fall; until then, eligible customers can use Fable 5.1 with zero data retention. AWS says EFS was built with Anthropic and will let eligible customers use Covered Models while keeping data in a cloud environment they control.

Safeguards are more precise, not gone. Anthropic says its newest cybersecurity safeguards block 60 percent fewer false positives than before, and Claude Code users can expect about 60 percent fewer cyber-safeguard interventions per session versus the previous Fable 5 rules. Fable 5.1 may now be used to discover software vulnerabilities, but not to develop exploits. Dual-use tasks—penetration testing, exploit generation, and binary-based vulnerability scanning—still redirect to Opus models. Life-sciences research and development queries on Fable 5.1 still go to Opus.

Mythos 5.1 is the same underlying model with more permissive safeguards for vetted people and organizations. Anthropic says it will be available through two trusted access programs: the Cyber Verification Program, which currently covers certain Opus- and Sonnet-class models and will include Mythos-class models in the near future, and a Life Sciences Verification Program developed in partnership with the U.S. government. The Verge reports that Mythos 5.1 is also launching in Project Glasswing.

One API change is aimed at distillation and will affect some custom loops. For new API accounts created from launch day onward, it is no longer possible to manually edit Claude’s prior context in a multi-turn conversation while preserving the transcript of Claude’s prior thinking. Existing accounts are not currently affected. Anthropic says it is rolling the change out gradually and that a small number of customers’ custom integrations will be affected.

If you already run Claude Code or a Fable 5 tool loop, try Fable 5.1 at the Claude Code High default, watch cache-read spend, and check whether your harness edits thinking transcripts or trips cyber and biology redirects. Teams that need Mythos-class cyber or biology headroom will have to apply for trusted access rather than swapping a model ID.

Cheaper cache reads

Claude Code, effort defaults, and benches

Covered Model, Mythos access, and harness gotchas

Sources