On November 9, 2026, Anthropic unveiled two new variants of its most advanced model: Claude Fable 5.2, publicly available with strengthened safeguards, and Claude Mythos 5.2, restricted to vetted organizations working in sensitive fields such as cybersecurity or biology. Both variants share the same underlying model and target the most demanding technical work, with Anthropic still recommending Claude Opus 5.5 for everyday coding tasks.
Built for long agentic workloads
The examples Anthropic highlighted give a sense of the ambition: a team of agents running an early build of Fable 5.2 migrated a production service from Python to Rust, and the service now starts roughly 26 times faster. An engineer also used the model to design a prototype operating system, coordinating more than 1,000 subagents over eight days until it booted on a laptop.
The benchmarks that matter
On the evaluation side, Anthropic highlights several scores, some of which it describes as the best its own models have ever recorded.
| Benchmark | Fable 5.2 | Comparison |
|---|---|---|
| FrontierCode 2.0 (Xhigh effort) | 49.1% | highest score Anthropic has tested |
| FrontierSWE v2 | 69.9% | 62.3% for Claude Opus 5.5 |
| OSWorld 2.1 (partial) | 89.3% | — |
| Humanity’s Last Exam (with tools) | 74.2% | — |
| RoboDojo Sim (Max effort) | 64.9% | — |
Anthropic also says that, inside its own company, over 90% of changes merged into its core codebase are now written by Claude, with human oversight.
Pricing and availability
The model is available under the identifier claude-fable-5-2 via the Anthropic API, as well as on Amazon Web Services, Google Cloud, and Microsoft Azure. Standard pricing is $8 per million input tokens and $40 per million output tokens. An optional fast mode, priced at $16 and $80 per million tokens, speeds up generation by up to 2.5x. Anthropic also reports outputs up to 30% faster than Fable 5.1 and Mythos 5.1.
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-fable-5-2",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Audit this Node.js module"}]
}'A team of agents running Fable 5.2 migrated a production service from Python to Rust, and it now starts roughly 26 times faster.
What it means for development teams
Fable 5.2 landing in GitHub Copilot on the same day as the announcement confirms a trend already visible across the industry: this generation of models is built for long, multi-step missions rather than simple code completion. Teams building CI/CD pipelines around GitHub Actions should also keep an eye on how GitHub now blocks pull_request_target by default, since agentic coding tools and workflow security increasingly intersect.
Key takeaways
Fable 5.2 and Mythos 5.2 continue the trajectory set by earlier versions: faster, cheaper-to-run models evaluated on tasks that increasingly resemble real software engineering work. Fable 5.2 shipping inside GitHub Copilot on announcement day suggests that integration with developers’ everyday tools remains Anthropic’s priority.
Benchmark figures published by model vendors should always be read with some caution, since they are chosen by the company publishing them. What stands out here is the shift in framing: Anthropic is no longer selling a completion assistant, but a collaborator capable of carrying a task for days without constant supervision. Whether that autonomy holds up outside demo scenarios remains to be seen — Simon Janvier.
Further reading
Primary source: Anthropic’s official announcement, November 9, 2026.
