Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9, 2026. The two share the same underlying model weights; Fable 5.2 carries additional cybersecurity and biology safeguards for general availability while Mythos 5.2 is restricted to vetted organizations. The models are 30% faster than their 5.1 predecessors, priced at $8 per million input tokens and $40 per million output tokens, with an optional fast mode up to 2.5 times faster.
On autonomous coding, Anthropic reported that over 90% of changes merged into its own codebase are now written by Claude, and that a team of agents running an earlier version of Fable 5.2 migrated a critical production service from Python to Rust, with the service starting about 26 times faster afterward. Fable 5.2 scored 41.3% on the Remote Labor Index, roughly double Opus 5.5 at 21.3%, GPT-6 Astra at 20.8%, and Fable 5.1 at 17.9%.
In mathematics, 1,000 Fable 5.2 agents fully formalized Fermat's Last Theorem in 30 hours, compared to 11 days for Anthropic's first attempt and only 26% completion by Opus 5.5 in 18 hours. In theoretical physics, Fable 5.2 completed a black hole formation calculation for small universes that physicists had formulated in 2003 but left unfinished, confirming that black holes form suddenly in small universes and extending a 1983 Hawking-Page result that applied only to large universes; a physicist verified the work. Anthropic also said the model solved a second open problem in theoretical physics related to holographic duality.
On hardware design, Fable 5.2 generated complete files for three working electronic devices from short prompts, including schematics, PCB layouts, 3D-printable enclosures, and firmware. Diode Computers manufactured all three devices directly from the model's output without manufacturing change orders. Earlier models had left PCB layouts 95% complete but unusable because they rarely fixed placements that blocked routing; Fable 5.2 reworked its designs without being asked, rerouting tracks dozens of times and reassigning chip pins. On Diode's internal PCB Gym benchmark, it scored 69% versus 25% for Fable 5 at roughly one-third the API cost.
On robotics, Fable 5.2 scored 64.9% on RoboDojo, more than double GPT-6 Astra and Fable 5.1. On safety, Mythos 5.2 used a planted sandbox-escape route in only 3.1% of scenarios, down from 39% for Mythos 5 and 21% for Mythos 5.1. Anthropic said Fable 5.2 has the strongest anti-distillation protections of any model it has released, reducing data harvested by distillers by 90% to 99%.
Anthropics documentation also shows that Fable 5.2's coding behavior shifts based on prompt framing: adding a sentence framing the codebase as the model's own caused it to merge repeated code in roughly half of test cases, versus fewer than one in ten without that framing.
Fable 5.2 launched on day zero in Cursor, where it scored 60% on CursorBench at max effort and 56.9% at default effort, up from 49.2% for Fable 5.1. Databricks reported it set a new high on OfficeQA Pro v2.