Claude Fable 5.1: Anthropic's New Model, Explained Simply

Anthropic released two new AI models on September 1, 2026, Claude Fable 5.1 and Claude Mythos 5.1, which it calls its most advanced yet for coding, research, and long-running problem-solving. Most people will only ever use Fable 5.1, so we explain what's actually different, what it costs, and how to try it.

Fable 5.1 vs. Mythos 5.1: same model, different guardrails

This is the detail most coverage glosses over: Fable 5.1 and Mythos 5.1 are the same underlying model, and the difference is in the safeguards. Fable 5.1 ships with standard safety restrictions and is generally available to everyone right now. Mythos 5.1 relaxes some of those restrictions specifically for vetted professionals in cybersecurity or life sciences, and it's only accessible through Anthropic's Cyber Verification Program or Life Sciences Verification Program, not something a regular user can sign up for.

What actually got better

Anthropic's own benchmark numbers show real jumps over the previous Fable 5:

• Scientific research tasks: 52.6% vs. 24.7% on Terminal-Bench-Science 0.1, more than double.
• Agentic coding: up to 60.9% vs. 42.0% on Terminal-Bench 4.0.
• General knowledge work: 1,853 vs. 1,723 on GDPval-AA v2.

Put simply, it's meaningfully better at multi-step coding tasks, research-style work, and staying coherent over long-running problems, not just a minor version bump.

It's also cheaper to run

Per-token pricing is unchanged at $10 per million input tokens and $50 per million output tokens. The real savings come from prompt caching: cache-read costs dropped 75%, to $0.25 per million tokens. Anthropic estimates this brings typical workload costs down about 25%, and up to 45% for applications that lean heavily on AI agents making repeated calls.

How to actually use it

Fable 5.1 is available immediately, with no waitlist, on Claude.ai, and through AWS, Google Cloud, and Microsoft Azure for developers. If you're calling it via the API directly, the model name is claude-fable-5-1. Mythos 5.1 is not something most people need to worry about unless they're specifically doing vetted cybersecurity or life-sciences work.

Under the hood, Fable 5.1 alters the standard sampling temperature dynamics. Where general reasoning models suffer increased hallucination and logical drift when temperature exceeds 0.7, Fable 5.1's underlying entropy calibration maintains syntactic coherence even under high variance sampling parameters, preventing repetitive tropes and formulaic sentence structures.

Should you switch to Fable 5.1?

If you use Claude through the app for writing, summaries or everyday questions, you will mostly get the new model automatically and there is nothing to configure. The benchmark gains matter most for long coding sessions, research-style tasks and agents that run for many steps. For short questions, the difference may be hard to notice.

How to test it on your own work

Published benchmarks are a starting point, not a verdict. A simple way to judge a new model is to take three or four tasks you already do, such as debugging a function, summarising a long document or drafting a report, and run the same prompts on the old and new model. Compare accuracy, how often you need to correct it and how long it stays on track. If you call the API, keep the model name in one setting so you can switch back easily.

Understanding the pricing change

Prompt caching stores a long, repeated prefix, such as a system prompt or a large document, so later requests reuse it at a lower rate. That is why a cut in cache-read cost helps applications that send the same context many times, such as coding agents, far more than a chat that rarely repeats itself. Your own savings will depend on how much of your traffic is cached. For how Anthropic's other models compare, see our Opus 5.5 explainer and the Sonnet 5.5 vs Opus 5.5 comparison.

Frequently asked questions

Related articles

Comments