Claude Fable 5.2 designed three working gadgets from a one-paragraph prompt

Anthropic‘s new Claude Fable 5.2 was given a single paragraph of instructions and came back with three electronic devices that worked: a home voice-assistant hub, a USB speakerphone and a handheld virtual pet. Claude wrote the specifications, picked real parts, drew the circuit boards, modelled the 3D-printed cases and programmed the software. It is the clearest sign yet that the company’s most capable model is meant to finish whole projects, not just answer questions about them.

For anyone paying for Claude, the upgrade arrives cheaper and quicker. Anthropic has cut the price of Fable to $8 per million input tokens, down from $10, and $40 per million output tokens, and says answers come back up to 30% faster than with Fable 5.1. As TechCrunch noted, that puts Fable below its main rival, OpenAI‘s GPT-6 Astra, on input price. Subscribers on eligible paid plans also get extra Fable usage during launch week.

The hardware test was run with Diode, a company that designs and manufactures circuit boards. According to Anthropic, Claude chose components available from US distributors, routed each four-layer board and had every design ready for manufacturing in under a day with minimal human input. Diode’s engineers built the boards and Claude guided them through testing and assembly. All three devices worked without any electrical modifications.

Physical tasks are a recurring theme in the release. Anthropic says Fable 5.2 controls robots far better than earlier Claude models, including on jobs it was never trained for, such as solving a puzzle cube, unscrewing a bottle cap and drawing a picture. On RoboDojo, a public benchmark in which an AI steers a simulated two-armed robot from camera images, Fable 5.2 scored 64.9% in Anthropic’s runs, more than double GPT-6 Astra at 29.8% and Fable 5.1 at 27.6%.

In software, Anthropic says more than 90% of the changes merged into its own codebase are now written by Claude under human guidance. A team of agents running an early version of Fable 5.2 moved a production service from Python to Rust, and it now starts about 26 times faster. On FrontierCode, a test of whether open-source maintainers would accept a model’s code changes, Fable 5.2 reached 49.1% at its highest effort setting, which Anthropic says is the best score of any released model it has tested. About 1,000 Fable 5.2 agents also formalised the proof of Fermat’s Last Theorem in the Lean language in 30 hours, a job an earlier research version of Claude needed 11 days to finish.

The caveats are real. Every benchmark figure above comes from Anthropic’s own runs, and the gadgets still needed people at Diode to build and assemble them. Anthropic itself admits a testing problem: Fable 5.2 and its sibling Mythos 5.2 voice suspicion that they are being evaluated more often than any of its recent models, which makes it harder to predict how they will behave in real use. Fable’s safety filters also block most cybersecurity work and most research biology, handing those requests to an older Opus model instead. Unless a business qualifies for Anthropic’s Enterprise Frontier Safeguards, its Fable data is kept for 30 days for safety monitoring. For everyday coding and writing, Anthropic still recommends starting with Claude Opus 5.5, which it calls its strongest daily driver, and moving up to Fable only for the hardest problems.

Fable and Mythos are the same underlying model. Fable carries the extra safeguards that let Anthropic offer it to everyone, while Mythos, which is better at finding software vulnerabilities, is limited to vetted organisations in Anthropic’s cyber and life sciences verification programmes. An optional fast mode generates text up to 2.5 times faster at double the price, $16 per million input tokens and $80 per million output tokens.

Claude Fable 5.2 went on sale on November 9, 2026, in the Claude apps for eligible paid plans and for developers on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure, in the countries where Anthropic offers Claude. Mainland China, Hong Kong and Russia are not among them. It replaces Fable 5.1, which arrived in September. The extra launch-week allowance gives subscribers one to three Fable resets depending on how long they have subscribed, and lets Fable use up to half of the weekly limit on Pro plans and standard Team seats.

Tags: artificial intelligence, Anthropic, Claude, AI models, Chi Pu, GPT-6 Astra