Claude Fable 5.1

What is disclosed

Claude Fable 5.1 came out on 1 September 2026.¹¹1Anthropic, Claude Fable 5.1 model page, Claude Platform docs, read 29 September 2026. https://platform.claude.com/docs/en/models/fable-5-1/overview Of the four models Anthropic lists as current, it has the highest token prices.²²2Anthropic, Claude API pricing, read 29 September 2026: Fable 5.1 at $10 input, $50 output and $0.25 for cache reads; Fable 5 at $10, $50 and $1; Opus 5.5 at $4 and $20. https://platform.claude.com/docs/en/about-claude/pricing. Arithmetic: 0.25 / 1 = 0.25, so cache reads cost 75% less; 4 / 10 = 20 / 50 = 0.4, so Opus 5.5 costs 60% less. Anthropic does not say how large it is or how it is built.

It is also sold under another name. Claude Mythos 5.1 is "the same model, but with different levels of safeguards".³³3Anthropic, Introducing Claude Fable 5.1 and Claude Mythos 5.1, September 2026. https://www.anthropic.com/claude-fable-and-mythos-5-1 Anyone can buy Fable 5.1. Mythos 5.1 goes by invitation only, to participants in Project Glasswing.¹¹

A request can carry 1M tokens of context and get back up to 128K. The model takes text and images and returns text, and Anthropic gives June 2026 as its reliable knowledge cutoff.¹¹ Thinking is adaptive and always on, at high effort unless the request says otherwise.¹¹

What is new

Input and output prices did not move from Claude Fable 5, at $10 and $50 per million tokens.²² Cache reads did. They fell from $1 to $0.25 per million, a cut of 75%.²² Anthropic estimates typical workloads at 25% cheaper than on Fable 5, and highly agentic ones at up to about 45% cheaper.³³

On Terminal-Bench 4.0, Fable 5.1 scores 55.8% and Mythos 5.1 scores 60.9%.³³ Anthropic puts the gap down to "earlier, less precise cyber safeguards" that stepped in on some Fable 5.1 tasks. Fable 5 had scored 42.0%, and the table gives no effort level.³³

The rules changed with the release. Fable 5.1 may now be used to find software vulnerabilities, though not to write exploits for them.³³ New API accounts can no longer edit earlier turns of a conversation while keeping the model's prior thinking. Anthropic calls the practice a common, publicly documented distillation technique.³³

Models released after 2 August 2026 put an invisible watermark in their outputs. Anthropic added it under the EU AI Act Code of Practice it signed in July 2026.³³

Using it

The id is claude-fable-5-1.¹¹ The models overview sends unsure users to Claude Opus 5.5 and keeps Fable 5.1 for "demanding reasoning and long-horizon agentic work".⁴⁴4Anthropic, Models overview, Claude Platform docs, read 29 September 2026. https://platform.claude.com/docs/en/about-claude/models/overview The Fable 5.1 page itself still says to start with Claude Opus 5.¹¹ Opus 5.5 costs 60% less per input and output token.²²

Three changes break code written for Fable 5. Forced tool use returns an error. Earlier models cannot read its thinking blocks, and editing earlier turns invalidates them.¹¹

Batch processing halves input and output prices.¹¹ A five-minute cache write costs $12.50 per million tokens.²² The docs promise availability until at least 1 September 2027.¹¹