Claude Fable 5.1

1:01 Claude Fable 5.1 is the most capable Claude model Anthropic offers to everyone, released on Sep 1, 2026 at $10 per million input tokens and $50 per million output tokens, the same price as Claude Fable 5 but with cache reads cut 75% to $0.25 per million. Anthropic positions it for demanding reasoning and long-horizon agentic work, and its docs suggest using it only when Claude Opus 5.5 at a higher effort level is not enough (Models overview, Anthropic docs, read Nov 6, 2026). It shares its weights with Claude Mythos 5.1, which runs with looser safeguards for verified security and life sciences organizations.

Quick facts

Sources: Claude Fable 5.1, Pricing and Model deprecations, Anthropic docs, read Nov 6, 2026; Introducing Claude Mythos and Fable 5.1, Anthropic, Sep 15, 2026; Plans and pricing, Claude, read Nov 6, 2026.

When was Claude Fable 5.1 released?

Anthropic's newsroom lists "Introducing Claude Mythos and Fable 5.1" on Sep 15, 2026 (Models, Anthropic, read Nov 6, 2026). The model's docs page says "Released September 1, 2026", the system card is dated Sep 1, and the Claude Code changelog added claude-fable-5-1 on Sep 1 (Claude Code changelog, Sep 1, 2026). Both dates come from Anthropic's own pages, but the post was public on Sep 1.

0:20 In Bijan Bowen's hands-on video of Sep 1, 2026, the Claude web app shows a "Claude Fable 5.1 is now available" notice, and the "Claude Fable 5.1 and Mythos 5.1" announcement appears on anthropic.com while he records. Anthropic posted its own launch video, Introducing Claude Fable 5.1, the same day. Sep 1 is when the model and its announcement reached readers; the newsroom's Sep 15 date is the odd one out.

What changed from Fable 5?

Anthropic lists four kinds of change in its announcement (Introducing Claude Mythos and Fable 5.1, Sep 15, 2026):

• Cost. The token price is unchanged, but cache reads dropped from $1 to $0.25 per million, and Anthropic estimates the total cost falls about 25% on typical work and up to about 45% on highly agentic work. It says Low or Medium effort gives results similar to or better than Fable 5's.
• Safeguards. Cyber safeguards intervene about 60% less per Claude Code session; Fable 5.1 may now help find vulnerabilities but not write exploits. Biology safeguards fire 85% less often on benign basic biology and medical questions.
• Privacy. Eligible customers can use zero data retention until Enterprise Frontier Safeguards, a customer-controlled storage option, arrive.
• Distillation. New API accounts can no longer edit earlier turns while keeping Claude's prior thinking.

For developers, the docs add breaking changes: forced tool use returns an error, editing earlier turns invalidates thinking blocks, and only Fable 5.1 and Mythos 5.1 can read Fable 5.1's thinking blocks. New beta features include changing effort mid-conversation, turn-scoped system messages and readable progress updates between tool calls (Claude Fable 5.1, read Nov 6, 2026).

Claude Fable 5.1 benchmarks

Anthropic's launch table, with its own runs of rival models:

Source: Introducing Claude Mythos and Fable 5.1, Anthropic, Sep 15, 2026. Mythos 5.1 scored 60.9% on Terminal-Bench 4.0 in the same table; Anthropic says the gap reflects where Fable's cyber safeguards stepped in, and tasks blocked by safeguards counted as zero.

A week later, Anthropic's Claude Opus 5.5 post put Fable 5.1 at 1735 on GDPval-AA v2.1, a newer version of the test whose scores are not comparable with v2, and at 65.6% on Humanity's Last Exam with tools rather than 65.0% (Introducing Claude Opus 5.5, Sep 22, 2026).

Independent results are less flattering. The public Terminal-Bench 4.0 leaderboard shows Fable 5.1 at 57.88% at xhigh effort, behind Opus 5.5 (64.85%) and Sonnet 5.5 (61.82%), read Nov 6, 2026. On Artificial Analysis's Intelligence Index its best score is 50, against 59 for Opus 5.5, at $7.54 per task at high effort (read Nov 6, 2026). On Arena's text leaderboard, Fable 5.1 at max effort ranks fifth with 1504 (table dated Oct 27, 2026). More on Claude Benchmarks.

What Anthropic says Fable 5.1 has done

Anthropic's post gives two science examples: Fable 5.1 trained a neural network that remapped about a third of Venus from Magellan radar data, sharpening detail from 10 to 20 km down to 2 to 3 km, and Mythos 5.1 designed protein binders with about ten times the affinity of the best entries in Adaptyv Bio's competitions on three targets (Introducing Claude Mythos and Fable 5.1, Sep 15, 2026). These are Anthropic's own claims and have not been independently reviewed on the page.

How to use Claude Fable 5.1

1. In the Claude apps: pick Fable 5.1 in the model menu next to the send button. On Pro it is paid with usage credits; on Max and Ultra it is included up to half of the weekly limit (Plans and pricing, read Nov 6, 2026). It does not work in Voice Mode.
2. In Claude Code: run /model fable or start with claude --model fable; the best alias also picks Fable where a plan has it (Model configuration, read Nov 6, 2026).
3. On the API: send "model": "claude-fable-5-1". It is on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS (Models overview).

2:02 Thinking cannot be switched off; effort is the control. Anthropic's advice for open-ended work is to start Fable 5.1 at medium effort and raise it as needed, and to stay on Opus or Sonnet for routine tasks.

Limits and known issues

Anthropic's own list: the model sometimes bypasses approvals and auto-mode classifiers, and its audits cover very long-context and multi-agent use only partly (Introducing Claude Mythos and Fable 5.1, Sep 15, 2026). Penetration testing and exploit generation requests still go to an Opus model instead. It is the slowest current model by Anthropic's own label, and at five times the price of Sonnet 5.5 it is the most expensive one to run.

What reviewers found

35:32 In a Sep 1, 2026 video, Bijan Bowen ran his usual coding tests on the Max plan at high and extra high effort. A self-contained C++ skateboarding game took an hour and ten minutes and needed one follow-up to fix its ollie, after which he called it "an absolute masterpiece"; a subway shooter built in Ultra Code took over three hours, two of them spent only on a design document, and he rated it "absolutely top tier". A 3D watch page left him unimpressed: "It's good, but it's not $50 per million." His tests ran past his plan's limits and cost $156 in extra usage, which he said shows how expensive these models are without subscription pricing.

0:00 Theo Browne of Theo - t3.gg called Fable 5.1 his "favorite code model ever released" in a Sep 29, 2026 video, but said he had not chosen Fable for any task in over a week, since Claude Opus 5.5 took over his workflow. He still rated Fable 5.1 the best of the three for good-looking front-end design.

Frequently asked questions

How much does Claude Fable 5.1 cost? On the Claude API, Claude Fable 5.1 costs $10 per million input tokens and $50 per million output tokens, half that through the Batch API. In the apps it needs Max, Ultra, a Team premium seat or Enterprise, or usage credits on Pro.

Is Claude Fable 5.1 better than Opus 5.5? Anthropic says Claude Opus 5.5 matches Fable 5.1 on most work, and its own Opus 5.5 table shows Opus 5.5 ahead on every row. Independent leaderboards read on Nov 6, 2026 also put Opus 5.5 ahead, so Fable 5.1 is worth its price mainly where your own tests show a gain.

Is Claude Fable 5.1 free? No. Claude Fable 5.1 is not on the Free plan; Pro subscribers pay for it with usage credits.

What is the difference between Fable 5.1 and Mythos 5.1? They are the same model. Claude Mythos 5.1 has looser safeguards for cybersecurity and life sciences work and is open only to verified organizations.

What is the Claude Fable 5.1 context window? Claude Fable 5.1 accepts 1M tokens of context and writes up to 128K tokens per response, in the apps on paid plans as well as on the API.

Sources

• Introducing Claude Mythos and Fable 5.1, Anthropic, Sep 15, 2026
• Claude Fable 5.1, Anthropic docs, read Nov 6, 2026
• Models overview, Anthropic docs, read Nov 6, 2026
• Pricing, Anthropic docs, read Nov 6, 2026
• Model deprecations, Anthropic docs, read Nov 6, 2026
• Models, Anthropic, read Nov 6, 2026
• Introducing Claude Opus 5.5, Anthropic, Sep 22, 2026
• Plans and pricing, Claude, read Nov 6, 2026
• Model configuration, Claude Code docs, read Nov 6, 2026
• Claude Code changelog, Anthropic, read Nov 6, 2026
• Terminal-Bench leaderboard, tbench.ai, read Nov 6, 2026
• Models leaderboard, Artificial Analysis, read Nov 6, 2026
• Text Arena, Arena, Oct 27, 2026
• Which Claude model should you use?, Claude, Sep 18, 2026 (video)
• Introducing Claude Fable 5.1, Anthropic, Sep 1, 2026 (video)
• Claude Fable 5.1 Is INSANE - Hands-On With the BEST Model Yet!, Bijan Bowen, Sep 1, 2026 (video)
• OpenAI should be scared of this one, Theo - t3.gg, Sep 29, 2026 (video)