Fable 5.2 solves physics problems and boosts freelance task completion

Anthropic highlighted the scientific impact of its Fable 5.2 model, noting that it solved two open problems in theoretical physics concerning black holes. The model achieved this by performing calculations that were previously too laborious for human scientists. In the field of biology, the related Mythos 5.2 model scored 85.1% on protein design benchmarks, a six-point improvement over its predecessor. It was used to design a new method for predicting the structure of protein capsids used in gene therapy, accurately modeling 28 of 34 attempted capsids. Mathematical capabilities also advanced, with Fable 5.2 fully formalizing Fermat's Last Theorem into a machine-readable proof. This task required 30 hours and 1,000 agents, whereas the previous Opus 5.5 model reached only 26% of the proof in 18 hours.

Beyond scientific research, the model's operational efficiency has improved significantly. Anthropic's official blog post indicates that the 'real freelance score' for Fable 5.2 increased from 21% to 41%. This metric reflects the model's ability to complete freelance tasks end-to-end. The system can now identify specific job postings, such as an equity analyst report for TDMX on Upwork, and execute the work for a fixed price of $450. This capability enables users to sell results as productized services rather than hourly labor. The model can also operate legacy software without APIs, such as exporting invoices from an old invoicing app and matching them against bank statements to identify unpaid items. Chabion X argued that while coding capabilities receive most attention, the model's work in science and engineering is accelerating problem-solving across fields, suggesting 2027 will see further rapid advancements.