What it is
Claude Fable 5.1 is Anthropic's most capable generally available model, released on September 1, 2026. It sits above Opus in the lineup and is meant for the hardest reasoning and very long agent tasks. It reads text and images, writes text, and has a 1M-token context with 128K output on the synchronous API.
Thinking is always on and defaults to high effort on the API. The price is $10 per million input tokens and $50 per million output tokens, with cache reads cut by 75% to $0.25. Claude Mythos 5.1 is the same model with looser safeguards, currently limited to a set of US organisations verified through Anthropic's cyber and life-sciences programmes.
It is available through the Claude API, the Claude apps, Amazon Bedrock, Google Cloud and Microsoft Foundry. Anthropic commits to keeping it available until at least September 1, 2027.
Our review
Fable 5.1 is an excellent model that Anthropic has already told most customers not to use. Its model page now says to start with Opus 5.5 and reach for Fable only when evaluations on Opus at higher effort still fall short. The reason is Opus 5.5, released on September 22, which by Anthropic's figures beats Fable 5.1 on Terminal-Bench 4.0 (66.4% against 55.8%), CursorBench 4.0 (57.8% against 51.8%) and GDPval-AA at 40% of the per-token price.
The margins are not uniform, and that is where Fable keeps a role. On Humanity's Last Exam with tools Opus 5.5 leads by only 2.1 points, and on OSWorld 2.1 by 1.1. Fable's launch figures show its biggest gain on scientific agent work: Terminal-Bench-Science rose from 24.7% on Fable 5 to 52.6%, though Anthropic gives a standard error of 3.5 to 4.5 points. Anthropic also says the gap to Opus in its internal use is narrower than the scores suggest.
The safeguards cost points and some workflows. Penetration testing, exploit generation and life-sciences research queries are handed to Opus, blocked tasks scored zero in Anthropic's own benchmarks, and the docs point developers to a refusals-and-fallback pattern. Anthropic says cyber interventions in Claude Code fell by about 60% from Fable 5, which helps, but a security team will still find the more useful model is Mythos 5.1, which it probably cannot get.
The cheap cache is the underrated change. At $0.25 per million, cached reads cost 2.5% of base input, against 5% on Opus 5.5, and Anthropic estimates long agent runs cost up to about 45% less than on Fable 5. Long, repetitive agent loops are where Fable's premium shrinks the most, though Opus 5.5 still wins on absolute cost.
Where it earns its place
A computational biology group running multi-day analysis agents on GPU code
Optimise model-training kernels and analysis pipelines that Opus has already failed to finish
Run Fable on the hardest jobs and Opus 5.5 on the rest. Anthropic reports GPU speed-ups of 1.4x to 7x on models such as Evo 2 and Enformer from this kind of work; life-sciences R&D prompts may still be routed to Opus by the safeguards.
An ML platform engineer building an agent harness for a 200-person research firm
Route only the hardest research tasks to the most capable model without rebuilding the stack
Pin claude-fable-5-1 behind an escalation rule and keep Opus 5.5 as default. At 1,000 long tasks a month, the bill is roughly $3,600 on Fable against about $1,500 on Opus, so the escalation threshold is the main cost lever.
A head of finance research at an investment bank
Decide whether Fable's quality on long financial analysis justifies its premium over Opus
Partner figures in the launch post (Samaya's FrontierFinance at 55.9% against 49.2% for Fable 5) are vendor-reported and predate Opus 5.5. A two-week evaluation on the bank's own tasks, with Batch pricing at half rate, is cheaper than guessing.
Before you commit
- Setup
- Same Messages API as other Claude models with model ID claude-fable-5-1; migrating from Fable 5 has four breaking changes, including errors on forced tool use and thinking blocks that only Fable 5.1 and Mythos 5.1 can read.
- Integrations
- Claude API, Claude apps, Claude Code, Amazon Bedrock (anthropic.claude-fable-5-1), Claude Platform on AWS, Google Cloud and Microsoft Foundry.
- Data and privacy
- Anthropic does not train on API inputs or outputs by default and deletes them within 30 days, with longer retention for flagged policy violations (up to two years) and limited retention required for safety work on covered models; zero data retention needs a separate agreement. Outputs carry an invisible watermark for EU AI Act purposes.
- Cost at scale
- An agent task with 500K input tokens (400K of them cached) and 50K output costs about $3.60 on Fable 5.1 against about $1.48 on Opus 5.5, before cache-write charges; Batch halves both for work that can wait.
- Limits
- 1M context, 128K output on the synchronous API; latency rated 'slower' than every other current Claude model. Rate limits are not given on the model page. Pen-testing, exploit generation and life-sciences R&D queries go to Opus, and Anthropic notes the model can still sometimes bypass approval and auto-mode classifiers.
What's good
- Largest jump in the lineup on scientific agent work (Terminal-Bench-Science 52.6%)
- Cache reads at $0.25 per million, 2.5% of base input
- 1M-token context, 128K output, retirement not before September 2027
What's not
- 2.5x Opus 5.5's per-token price, and behind it on Anthropic's own benchmarks
- Slowest model in the Claude lineup
- Security and life-sciences requests rerouted to Opus; forced tool use returns an error
Best for
Teams with long scientific or research agent tasks where Opus 5.5 at high effort has been tested and found short.
Alternatives
Claude Fable 5.1: common questions
- Is Claude Fable 5.1 worth it?
- Three weeks after launch, Anthropic's own cheaper Opus 5.5 overtook Fable 5.1 on most published benchmarks, which leaves it a specialist's model at a generalist's flagship price. We rate it 4.3 out of 5.
- How much does Claude Fable 5.1 cost?
- API: $10 per 1M input tokens, $50 per 1M output tokens; cache writes $12.50 (5 min) or $20 (1 hour); cache reads $0.25; Batch API 50% off.
- What are the downsides of Claude Fable 5.1?
- 2.5x Opus 5.5's per-token price, and behind it on Anthropic's own benchmarks; slowest model in the Claude lineup; security and life-sciences requests rerouted to Opus; forced tool use returns an error.
- Who is Claude Fable 5.1 best for?
- Teams with long scientific or research agent tasks where Opus 5.5 at high effort has been tested and found short.
- How much does Claude Fable 5.1 cost for a team or in production?
- An agent task with 500K input tokens (400K of them cached) and 50K output costs about $3.60 on Fable 5.1 against about $1.48 on Opus 5.5, before cache-write charges; Batch halves both for work that can wait.
- Is Claude Fable 5.1 safe for company data?
- Anthropic does not train on API inputs or outputs by default and deletes them within 30 days, with longer retention for flagged policy violations (up to two years) and limited retention required for safety work on covered models; zero data retention needs a separate agreement. Outputs carry an invisible watermark for EU AI Act purposes.
- What are the best Claude Fable 5.1 alternatives?
- The closest alternatives we have reviewed are Claude Opus 5.5, GPT-6 Astra and Gemini 4 Argon.
Answers checked Oct 22, 2026.
Make Claude Fable 5.1? Show this rating on your site
