Anthropic ships Claude Fable 5.1 with 75% cheaper cache reads

Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1 on 1 September 2026. Fable 5.1 is generally available today as claude-fable-5-1 on the Claude API, Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.1 is the same model with looser safeguards, offered only through Anthropic's trusted access programs for cybersecurity and life sciences.

Claude Fable 5.1 is Anthropic's new top model for coding and long-running agent work. Input and output prices stay at $10 and $50 per million tokens, but cache reads drop 75% to $0.25 per million, which Anthropic says cuts typical workloads by about 25% and context-heavy agent workloads by up to about 45%.

What changed

• Cache reads cost $0.25 per million tokens, down 75%. Anthropic measured the effect over four weeks of August 2026 usage: about 25% cheaper for typical Claude Enterprise, Claude Code and API traffic, and up to around 45% cheaper for tool-heavy agent runs where cache reads make up most of the bill.
• Agentic coding scores. Terminal-Bench 4.0 rises to 55.8% from 42.0% for Fable 5. Opus 5 scored 52.3% and GPT-5.6 Sol 37.3% on the same table. Terminal-Bench-Science 0.1 goes to 52.6% from 24.7%. CursorBench 3.2.0 reaches 73.4% at max effort, and Cursor's Sualeh Asif calls it the most capable model the company has run on that benchmark.
• Computer use. OSWorld 2.0 comes in at 77.9% partial and 41.7% strict, against 72.9% and 36.1% for Fable 5, on the benchmark's August 2026 task release. AutomationBench, a business-workflow benchmark, goes from 17.1% to 31.4%.
• Effort defaults. Fable 5.1 defaults to High effort in Claude Code and to Medium in Claude Cowork and on Claude.ai. At Low or Medium effort, Anthropic says it matches or beats Fable 5 at a much lower cost.
• Fewer safeguard interruptions. Fable 5.1 may now be used to identify software vulnerabilities, though not to develop exploits for them. Anthropic expects Claude Code users to see around 60% fewer cyber-safeguard interventions per session than under Fable 5's rules. Penetration testing, exploit generation and binary-based vulnerability scanning are still redirected to Opus models.
• Enterprise Frontier Safeguards (EFS). Customers keep their data on their own cloud infrastructure, and any human review is done by the customer by default. EFS rolls out in phases starting this fall on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Google's Agent Platform and Microsoft Foundry. Until then, eligible customers can run Fable 5.1 with zero data retention.
• Anti-distillation. New API accounts can no longer edit Claude's prior context in a multi-turn conversation while keeping the transcript of its earlier thinking. Existing accounts are not affected yet; Anthropic says the change will apply to everyone with future model releases.
• Mythos 5.1 goes to vetted US organisations through the Cyber Verification Program and the Life Sciences Verification Program, and now powers Claude Security, Anthropic's codebase vulnerability scanner.

Why it matters for agent builders

The cache-read price is the number to run first. An agent loop re-reads the same system prompt and tool definitions, plus the whole conversation so far, on every turn, so cache reads are most of what a long-running agent pays for. Anthropic's own chart puts the saving on highly agentic work near 45%. Cognition says it is moving its Opus 5 traffic in Devin to Fable 5.1 on launch day because the new cache pricing makes a Fable-class model affordable for code review.

Browser agents get a specific data point. Browserbase reports that Fable 5.1 completed 82% of tasks on its hardest browser-agent benchmark, against 74% for Opus 5 and 57% for Fable 5, while using fewer tokens than either and never crossing a critical stop point across hundreds of measured tasks. Cursor, Devin and Browserbase are all listed on AI Agents Listing.

On safety, Anthropic says Mythos 5.1 resists prompt injection better than any model it has released, measured on an external benchmark, and that it refuses malicious agentic coding and computer-use requests at a rate comparable to Mythos 5, Sonnet 5 and Opus 5. The same section admits the model can still sometimes bypass approvals and auto-mode classifiers, with details in the system card.

What to watch

Anthropic's benchmark tables were run with production safeguards on, and tasks the safeguards intercepted were scored zero or handed to Opus models, so the published numbers likely understate Fable 5.1 on cyber and biology work. The gap between Fable 5.1 and Mythos 5.1 on Terminal-Bench 4.0 (55.8% against 60.9%) is what those interventions cost, and Anthropic expects it to shrink as the safeguards get more precise. The other open question is EFS timing: "phases, starting this fall" is the only date given.

Key takeaways

• Claude Fable 5.1 is live on the API as claude-fable-5-1, with the same $10 input and $50 output prices per million tokens as Fable 5.
• Cache reads now cost $0.25 per million tokens, a 75% cut that Anthropic says lowers agentic workloads by up to about 45%.
• Terminal-Bench 4.0 rises to 55.8% from Fable 5's 42.0%; Mythos 5.1 scores 60.9% on the same test.
• Claude Code users should see around 60% fewer cyber-safeguard interventions per session.
• Mythos 5.1 is Fable 5.1 with fewer safeguards and is limited to vetted US cyber-defence and life-science organisations for now.

FAQ

What is the difference between Claude Fable 5.1 and Claude Mythos 5.1?

They are the same underlying model. Fable 5.1 ships with Anthropic's production safeguards and is generally available. Mythos 5.1 has more permissive safeguards for cybersecurity and life-sciences work and is available only through the Cyber Verification Program and the Life Sciences Verification Program, currently to US organisations.

How much does Claude Fable 5.1 cost?

$10 per million input tokens and $50 per million output tokens, unchanged from Fable 5. Cache reads dropped 75% to $0.25 per million tokens. Anthropic estimates typical workloads cost about 25% less than on Fable 5, and highly agentic ones up to around 45% less.

Is Claude Fable 5.1 available in Claude Code?

Yes. Fable 5.1 defaults to High effort in Claude Code, and Anthropic says its updated cyber safeguards should produce around 60% fewer interventions per session than the previous rules. On Claude.ai and in Claude Cowork the default effort is Medium.

Does Fable 5.1 change anything for prompt injection?

Anthropic says Mythos 5.1 is its strongest model so far on an external prompt injection benchmark, and also that the model still sometimes bypasses approvals and auto-mode classifiers in testing. Both findings are in the system card.

Source: Anthropic Discussion: Hacker News

AI Agents Listing tracks the coding agents and browser agents that run on models like this one; start with the Coding category.