Claude Fable 5.1

// Overview

Claude Fable 5.1 is Anthropic's Mythos-class model released on 1 September 2026, built for demanding reasoning and long-horizon agentic work. It carries a 1M-token context window and up to 128K output tokens, accepts text and image input, and runs adaptive thinking that is always on, with an effort parameter that defaults to high. The API id is claude-fable-5-1 on the Claude API, Amazon Bedrock (anthropic.claude-fable-5-1), Google Cloud, Microsoft Foundry, and Claude Platform on AWS.

Pricing holds at Claude Fable 5's $10 per million input tokens and $50 per million output tokens, but cache reads fall 75% to $0.25 per million. Anthropic reports that this works out to roughly 25% lower cost on typical workloads and up to around 45% on complex coding and highly agentic tasks, where the same cached context is read back on every step. A 5-minute cache write costs $12.50 per MTok, a 1-hour cache write $20 per MTok, and the Batch API takes 50% off input and output.

Anthropic's launch table puts Claude Fable 5.1 at 55.8% on Terminal-Bench 4.0 against 42.0% for Claude Fable 5 and 52.3% for Claude Opus 5, and at 52.6% on Terminal-Bench-Science 0.1 against 24.7% for Fable 5. It reports 60.9% on Humanity's Last Exam without tools, 73.4% on CursorBench 3.2.0, 77.9% on the partial scoring of OSWorld 2.0, and 31.4% on AutomationBench. Anthropic also reports that Claude Mythos 5.1 designed protein binders with binding affinities 10 times higher than the best entries in Adaptyv Bio's protein design competitions on three targets, and sped up seven open-source deep learning models by up to 2.5 times with custom GPU kernels.

Three API behaviours are breaking changes for callers moving off Claude Fable 5: forced tool use returns an error, earlier Claude models cannot read Fable 5.1's thinking blocks, and editing an earlier turn invalidates thinking blocks. Five additions are optional — per-message effort (beta), turn-scoped system messages (beta), readable progress updates between tool calls (beta), the lower cache read price, and content provenance. Claude Mythos 5.1 is the same underlying model with lighter safeguards, offered by invitation only through Project Glasswing.

// Benchmarks

This model's scores

1. Terminal-Bench 4.0 (agentic coding)55.8%
2. Terminal-Bench-Science 0.1 (agentic scientific research)52.6%
3. CursorBench 3.2.0 (agentic coding)73.4%
4. Humanity's Last Exam (no tools)60.9%
5. OSWorld 2.0 (computer use, partial)77.9%
6. AutomationBench (business workflows)31.4%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

5-minute cache write $12.50/MTok; 1-hour cache write $20/MTok; Batch API 50% off input and output. Claude Mythos 5.1 shares this pricing.

Pricing source ↗

// Strengths

• Agentic scientific research — 52.6% on Terminal-Bench-Science 0.1, against 24.7% for Claude Fable 5 and 29.0% for Claude Opus 5
• Agentic coding — 55.8% on Terminal-Bench 4.0 and 73.4% on CursorBench 3.2.0
• Cache reads at $0.25 per MTok, a 75% cut that lands on long agent runs that re-read the same context
• 1M-token context window with up to 128K output tokens, and a June 2026 knowledge cutoff
• Per-message effort (beta) changes reasoning depth mid-conversation without invalidating the prompt cache

// Best for

• Run long-horizon agentic coding across large codebases where Claude Opus 5 at higher effort still falls short on your evals.
• Drive multistep research and terminal agents that keep a large cached context alive across many turns.
• Work through documents, spreadsheets and slides that fit a 1M-token context window.
• Handle vision-heavy analysis where the model must read figures and screenshots alongside text.

// How to access

// Claude Fable / Mythos (Mythos-class) — every version

The full lineage of the Claude Fable / Mythos (Mythos-class) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.