Anthropic released Claude Fable 5.2. It is designed to respond faster and costs a fifth less than its predecessor.

The most powerful publicly available Claude has received a successor. On November 9, Anthropic released Claude Fable 5.2, along with Mythos 5.2, which is intended only for vetted organizations. The new model is expected to be better at programming, reasoning, and image processing. However, the main news is the price: Fable 5.2 is cheaper than its predecessor.

A fifth cheaper, fast mode for an extra charge

According to Anthropic’s announcement, Fable 5.2 in the API costs 8 dollars per million input tokens and 40 dollars per million output tokens. Fable 5.1 cost 10 and 50 dollars. Those who need answers even faster can enable fast mode, which speeds up generation by up to 2.5 times. However, they will pay double for it, i.e., 16 and 80 dollars.

The context of 1 million tokens, responses up to 128 thousand tokens, and knowledge up to June 2026 remain the same as with Fable 5.1. Fable and Mythos are, in fact, one model. Fable also has safeguards for cybersecurity and biology, so risky queries will be answered by Opus 4.8 or Opus 5 instead. According to the company’s initial tests, unnecessary blockages are expected to be about 55% fewer than with Fable 5.1.

Mostly outperforms Opus 5.5, but not always Gemini

In standalone programming FrontierSWE v2, Anthropic reports 69.9% compared to 62.3% for Opus 5.5 and 65.5% for GPT-6 Astra. In computer control OSWorld 2.1, it achieved 89.3%. However, in the AutomationBench business process test, Gemini 4 Carbon surpassed it with 53.9%.

In the FrontierCode 2.0 programming test, Fable 5.2 with high effort is expected to score roughly the same as Opus 5.5 with maximum effort, i.e., 45.3% versus 45.5%, and that is for approximately half the cost per task. However, all figures come from the manufacturer; independent measurements are still lacking.

Fermat in 30 hours, three devices, and a custom system

A team of approximately a thousand Fable 5.2 agents converted the proof of Fermat’s Last Theorem into a form that a computer can verify step-by-step in 30 hours. The first attempt with an unreleased research version of Claude took 11 days.

According to Anthropic, the model also designed three electronic devices, including boards, enclosures, and software, and all worked without circuit modifications. In simulated robotic arm control, it achieved more than double the score of Fable 5.1. Additionally, one engineer had it write an operating system prototype: over a thousand sub-agents worked on it for eight days, and the result boots on a laptop.

🧪 Where to try Fable 5.2In the Claude application, you can find it in paid plans with access to Fable models. In the week of November 9, Pro subscribers and standard seats in the Team plan can use up to half of their weekly limit on Fable. In applications and Claude Code, it runs with medium effort. Developers can find it in the API as claude-fable-5-2 and also on AWS, Google Cloud, and Microsoft Azure.

However, for most programming and office work, Anthropic continues to recommend starting with Opus 5.5, which costs half as much.

Would you opt for Fable 5.2 for demanding tasks, or is the cheaper Opus 5.5 sufficient for you?

Sources: anthropic.com, platform.claude.com