Anthropic releases Claude Fable 5.2 and Mythos 5.2, cuts Fable price by 20 percent

Anthropic released two new models on 9 November 2026: Claude Fable 5.2 and Claude Mythos 5.2. The company describes them as built for "the most daunting technical challenges". It reports gains in coding, reasoning and vision, and says outputs come up to 30 percent faster than with the 5.1 versions released in September.

Fable 5.2 is generally available now. Mythos 5.2 goes only to vetted organizations. Fable's price drops to $8 per million input tokens and $40 per million output tokens, from $10 and $50. Anthropic's own examples come from a single place: long tasks that run for many hours or days, with many agents working in parallel and little human input.

What Anthropic says the model did

In the announcement, Anthropic lists results from prerelease testing. Agents running an earlier version of Fable 5.2 moved a critical production service from Python to Rust. Anthropic says the service now starts about 26 times faster. It also says its engineers used the model to finish migrations that were "otherwise infeasible".

The hardware example is specific. Fable 5.2 designed three working electronic devices from scratch: a home voice-assistant hub, a USB speakerphone and a handheld virtual pet. Each has its own four-layer circuit board, a 3D-printed enclosure and software. The model wrote the specifications, picked components from US distributors and modeled the enclosures in CAD. According to Anthropic, each board was ready for manufacturing in under a day. The board maker Diode built them, and Claude guided Diode's engineers through testing and assembly. All three devices worked without electrical modifications.

In physics, Anthropic says Fable 5.2 solved two open problems tied to holographic duality and black holes. One was a 2003 calculation, previously too laborious to finish, showing that black holes form abruptly in very small universes as they do in large ones. The other was finding the exact limit of a technique physicists have used for more than 15 years to rule out candidate theories of quantum gravity.

The most precise claim is in mathematics. A team of 1,000 Fable 5.2 agents fully formalized the proof of Fermat's Last Theorem in the Lean proof language in 30 hours, with minimal human intervention. An unreleased research model did the first formalization, which took 11 days with a purpose-built harness and human guidance. Anthropic says Fable 5.2 started from minimal prompts, built its own coordination methods and used no mathematical references. Its proof is seven times shorter. Claude Opus 5.5, given the same setup, had finished 26 percent after 18 hours. Anthropic notes that each model was run once.

In another experiment, Fable 5.2 started from an empty folder and coordinated more than 1,000 subagents around the clock for eight days. The result was a prototype operating system that boots on a laptop, with drivers, file storage, windows, a login screen and an installer.

The numbers

The product page shows Anthropic's benchmark table. On FrontierCode 2.0, an agentic coding test, Fable 5.2 scores 49.1 percent. Fable 5.1 scored 32.4 percent, Opus 5.5 scores 45.5 percent and OpenAI's GPT-6 Astra 42.8 percent. On the Remote Labor Index, which measures paid remote work tasks, Fable 5.2 reaches 41.3 percent against 17.9 percent for Fable 5.1. On the RoboDojo robotics simulation it scores 64.9 percent, up from 27.6 percent. The model does not lead everywhere. On AutomationBench, a business workflow test, Google's Gemini 4 Carbon scores 53.9 percent to Fable 5.2's 51.4 percent. These figures come from Anthropic, not from independent testing.

One customer quoted on the page reports that phonon calculation code in Quantum ESPRESSO ran about 36 times faster on average, falling from 3.3 hours to under 4 minutes across 13 cases. Another describes a 10 to 12 hour security audit run with about 170 reviewer agents.

Access and safeguards

Fable 5.2 is on Anthropic's paid plans, the Claude API as claude-fable-5-2, AWS, Google Cloud and Microsoft's cloud. A fast mode that generates up to 2.5 times faster costs $16 and $80 per million tokens. Mythos 5.2 is available through Anthropic's Cyber and Life Sciences Verification Programs.

Fable still routes flagged requests to less capable models: cyber queries go to Opus 4.8, biology queries to Opus 5. Anthropic says blocks are down about 55 percent compared with Fable 5.1. By design, the model's ability in AI research tasks such as accelerator kernel optimization stays at about the level of Opus 5.5. On Anthropic's CVE Rediscovery evaluation, Mythos 5.2 found 14.4 percent more known vulnerabilities than Mythos 5.1. Mozilla says it roughly doubled Mythos 5.1's vulnerability discovery metrics.

The release comes as Anthropic prepares an initial public offering. CNBC reports it could happen as soon as next week. Dianne Penn, Anthropic's head of product management for research, told CNBC that Fable 5.2 is aimed at "power users or power use cases". She said Opus 5.5 remains better suited to everyday tasks.

What it means for work

Anthropic's examples are the kind of work engineering teams usually put off because it takes too long: porting a production service to another language, formalizing a large proof, getting a circuit board ready for manufacturing. In these cases the model writes the specification, splits the work across hundreds or thousands of agents and checks the result. A person sets the goal at the start, intervenes now and then, and verifies at the end. For engineers, that moves the scarce skill from writing code to defining tasks precisely and judging output that no one has read line by line. Lower prices and faster output make multi-day agent runs cheaper to try. Still open: how these results hold up outside Anthropic's own tests, which ran once in several cases, and who is responsible when a thousand-agent run produces something subtly wrong.