Anthropic announced on Monday, November 9, 2026, two new models: Claude Fable 5.2 and Claude Mythos 5.2. The company describes them as its newest models for complex technical work.
The two models are not identical in availability. Fable 5.2 is available to all paid plans, and through the Claude Platform, Claude Code, and major clouds. Mythos 5.2 is available only for verified organisations in Anthropic's Cyber and Life Sciences Verification Programs. That distinction reflects the different kinds of work the two models are intended for, and the controls Anthropic places around them.
Pricing for Fable 5.2 is $8 per million input tokens and $40 per million output tokens, down from $10 and $50. Responses are up to 30 percent faster. Lower per-token pricing and faster responses both reduce the cost of running a model repeatedly, which matters for developers building agents and tools on top of it. Fable 5.2 also includes safeguards blocking dangerous cybersecurity and biology requests.
What Fable 5.2 Was Tested On
Anthropic reported a series of test results for Fable 5.2 that focus on long-running, multi-agent work, rather than single short answers.
In one test, Fable 5.2 coordinated more than 1,000 AI agents over eight days to build a working operating system from scratch that booted on a laptop, with windows, file storage, and a login screen.
In another test, a team of 1,000 Fable 5.2 agents formalised Fermat's Last Theorem in 30 hours. Formalisation means translating a mathematical proof into a form that can be checked mechanically. Anthropic reported that Opus 5.5, an earlier model, managed 26 percent of the same task in 18 hours.
Fable 5.2 also helped solve two previously open problems in black hole physics, and designed three working electronic devices. These tests share a pattern: many agents working together over a long period, on a task too large for a single model response. That is the direction of travel for agent-based software, where a system of agents carries out a project, coordinating work, checking results, and correcting course.
Mythos 5.2 and the Life Sciences
Mythos 5.2 is the more restricted of the two models. It is available only to verified organisations in Anthropic's Cyber and Life Sciences Verification Programs. In testing, Mythos 5.2 modelled 28 of 34 gene-therapy capsids. A capsid is the protein shell of a virus, and gene therapy uses modified viruses to deliver genetic material. Models that can work on cybersecurity and biology raise dual-use concerns: the same capability that can help a legitimate researcher can, in the wrong hands, be used to cause harm. Limiting access to verified organisations is one way to manage that risk while still allowing research to proceed.
Benchmarks and Ranking
Artificial Analysis put Fable 5.2 in its maximum configuration first on its Intelligence Index version 5, with a score of 65. That placed it five points ahead of Gemini 4 Carbon and six points ahead of Opus 5.5. A benchmark index is a composite score across a set of tests, designed to compare models on a common basis. A first-place ranking does not mean a model is best for every task, but it is a signal developers and buyers will notice.
The pricing change is also a competitive signal. Cutting input pricing from $10 to $8 per million tokens and output pricing from $50 to $40, while making responses up to 30 percent faster, lowers the cost of high-volume work. For a developer running agents that make many model calls, that difference compounds quickly.
Why It Matters for Founders and Investors
For founders building with AI, the launch matters in three ways. First, the multi-agent results suggest the practical frontier is shifting from single-model quality to coordination: how well a system can keep 1,000 agents working on one project for eight days. Products that help developers orchestrate, monitor, and debug those systems will become more valuable as the underlying models improve.
Second, the split between Fable and Mythos shows how access controls are becoming a product feature. Founders in cybersecurity and life sciences need to plan for verification, approval, and restricted access, rather than assuming they can use the newest model on day one.
Third, the pricing and speed changes affect unit economics. If a product makes many model calls per user, a 20 percent cut in per-token pricing and a faster response time can change the margin on each user.
For investors, the announcement is a reminder that the frontier model race is still moving quickly, and that the competition is now measured in agent coordination and scientific problem-solving as well as in chat quality. The companies that win developer adoption will be those that combine a strong model with clear pricing, reliable access, and safeguards that enterprises can accept.
The Daily Fledgling, November 10, 2026
Precise Impact AI, November 10, 2026
Read Next
OpenAI Releases Codex Security as Open-Source CLI and SDK
Databricks Unveils Always-On Genie, an AI Agent That Stops Waiting to Be Asked
DeepMind Launches AlphaGenome 2, a DNA Model That Reads Genomes Cell by Cell
ElevenLabs Unveils 13 New Products and Upgrades at Its New York Summit
Google Launches Gemini Decision 1, a Model That Answers Instead of Writing
Runway Open-Sources Praxis-1, Its Robot Model That Learned by Watching Video



