Claude Fable 5.1 and Claude Mythos 5.1

Overview

Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026, calling them the world’s most advanced models for coding and knowledge work. Both share the same underlying capabilities and differ only in safeguards — Fable 5.1 is generally available, while Mythos 5.1 is offered to Project Glasswing participants only. Input and output pricing is unchanged from Claude Fable 5 at $10 / $50 per MTok, with cache reads at a quarter of the previous cost. The model IDs are claude-fable-5-1 and claude-mythos-5-1.

Key features

• Stronger long-running agentic work

  The gains concentrate in six areas: agentic coding over long sessions, including multi-file features, large refactors and migrations, debugging, and code review across sessions that run for hours; knowledge work with documents, spreadsheets, and slides, taking an analysis from a first question to a finished deliverable built from a blank page; research and search across multistep and deep-research tasks; vision, reading dense charts, filings, and tables nested in PDFs; long-context work across the full 1M token context window; and computer use, operating a browser and desktop applications more reliably. The gap over Claude Fable 5 is widest at higher effort levels.

• Cache reads at a quarter of the cost

  Long agentic sessions re-read the same cached prefix over and over, so cache read cost accumulates. On Claude Fable 5.1 and Claude Mythos 5.1, cache reads (hits and refreshes) cost 0.025 times the base input price — $0.25 per MTok — compared with 0.1 on other Claude models. Base input and output prices are unchanged, yet Fable 5.1 costs an estimated 25% less than Fable 5 for typical workloads, and up to approximately 45% less on complex agentic tasks. Cache writes and the 512-token minimum cacheable prompt length are unchanged.

• Change effort mid-conversation (beta)

  Holding one effort level for a whole session meant paying for depth on routine steps or going without it on hard ones. On Claude Fable 5.1 you can change the effort level mid-conversation without invalidating the prompt cache — raise it for a hard step and lower it for routine ones. Per-message effort is in beta: include the mid-conversation-output-config-2026-07-01 beta header. Claude Fable 5.1, Claude Mythos 5.1, and Claude Opus 5 support it on the Claude API.

• Turn-scoped system messages (beta)

  Per-turn reminders in a tool loop used to mean injecting text into the history and deleting it on the next request, which edits the conversation and breaks both the cache and thinking blocks. Set clear_at: "next_user_message" on a role: "system" message and its text carries system-prompt authority for the current turn, then stops rendering once a later user message exists. The message stays in messages and you keep sending it back verbatim, so nothing earlier changes, and a cleared message costs no input tokens. Include the mid-conversation-system-clear-at-2026-08-21 beta header.

• Progress updates between tool calls (beta)

  Claude Fable 5.1 writes short progress updates between tool calls on what it found and what it will do next, but each arrives as a thinking block, and under the default thinking.display of "omitted" those blocks come back empty — so a long agentic turn can look silent to your users. The new display: "updates" option, set with the thinking-display-updates-2026-08-18 beta header, returns the progress updates as text while reasoning stays hidden.

• Scientific research capabilities

  Anthropic tested the models’ research capabilities across a range of domains. In molecular design, Mythos 5.1 designed binders with binding affinities 10 times higher than the best designs submitted to protein design competitions, reaching a hit rate of nearly 50% across 12 targets, against a typical 10-15%. The models also reconstructed a Venus elevation map revealing details down to two to three kilometers rather than 10 to 20, and accelerated GPU inference on deep learning models by up to 2.5 times.

• Fewer safeguard false positives

  Anthropic reports 60% fewer false positives than before in cybersecurity contexts: Fable 5.1 can identify software vulnerabilities for defensive work while still blocking exploit generation and penetration testing tasks. Biology safeguards fire 85% less often for benign requests such as elementary biology and medical questions. Research-level biology capabilities remain restricted to Mythos 5.1, through a partnership with the US government.

• Content provenance

  Text generated by Claude Fable 5.1 and Claude Mythos 5.1 carries Anthropic’s statistical text watermark on every platform where the model is available. The watermark doesn’t change the meaning, quality, or readability of the output, adds no tokens or hidden characters, and carries no information about you or your organization. Supported image and video files retrieved through the Files API carry signed C2PA Content Credentials.

Benchmarks

Mythos 5.1 scores 60.9% on Terminal-Bench 4.0.

Pricing & availability

Notes

• Three breaking changes when migrating from Claude Fable 5 — forced tool use is not supported, so tool_choice set to {"type": "any"} or {"type": "tool", "name": "..."} returns a 400 invalid_request_error (use strict tool use or structured outputs for schema-valid JSON). Earlier models can’t read Claude Fable 5.1 thinking blocks, so a conversation that moves from Fable 5.1 back to an earlier model loses its reasoning for those turns. And modifying the system prompt, the tools, or an earlier message invalidates every later thinking block.
• Treat the conversation as append-only — the history check is enforced for accounts created on or after August 31, 2026. Claude Code, claude.ai, Claude Managed Agents, and the Claude Agent SDK keep the prefix intact for you; code that builds the messages array itself needs the history-editing check before migrating. Add instructions with mid-conversation system messages and trim context with server-side context editing or compaction, which don’t count as edits.
• Default behavior differs from Claude Fable 5 — parallel tool calling is more variable, with one tool call per turn where Fable 5 batched several; there are fewer progress updates during long tool runs; at low effort it answers from memory more often instead of calling a search tool; prose is denser in places with less formatting in chat; summaries are more likely to reproduce source passages without marking them as quotations; and it is more likely to rewrite a whole file for a small edit. Each has a prompting fix in “Prompting Claude Fable 5.1”.
• Unchanged from Claude Fable 5 — adaptive thinking is always on, and both thinking: {"type": "enabled"} with budget_tokens and thinking: {"type": "disabled"} return a 400 error. thinking.display defaults to "omitted" and the raw chain of thought is never returned. Prefilling the assistant response and non-default temperature, top_p, or top_k values return a 400 error.
• Refusals, fallback, and billing — a declined request returns HTTP 200 with stop_reason: "refusal" and a stop_details object naming the policy area that fired. The permitted fallback targets for Claude Fable 5.1 are Claude Opus 4.8 and Claude Opus 5. You aren’t billed for a refusal that arrives before any output, and fallback credit refunds the prompt-cache cost of switching models.
• Data retention — both models are Covered Models carrying 30-day data retention, and aren’t available under zero data retention unless expressly authorized by Anthropic.
• Enterprise Frontier Safeguards has not shipped yet — it lets customers store data on their own cloud infrastructure rather than Anthropic systems while maintaining misuse detection, and rolls out in phases beginning fall 2026.