Claude Fable 5.2 and Mythos 5.2 Launch
Anthropic released both on Nov. 9. They share one underlying model: Fable has cyber and bio safeguards and is open to everyone, while Mythos is limited to vetted organizations. Both have a 1M context window, 128k max output and run up to 30% faster than the 5.1 versions.
Pricing drops to $8 per million input tokens and $40 per million output, from $10 and $50. Fast mode runs up to 2.5x faster at $16/$80. Anthropic still recommends Opus 5.5 as the default for most work.
Anthropic Says Fable 5.2 Beats Opus 5.5 and GPT-6 Astra
Anthropic's table lists Fable 5.2 at 49.1% on FrontierCode 2.0 (Opus 5.5: 45.5%), 69.9% on FrontierSWE v2 and 89.3% on OSWorld 2.1. It also cites a prototype operating system built by over 1,000 subagents in eight days.
These are vendor-reported numbers. Treat them as a starting point until independent testers weigh in.
New Breaking Changes in the API
The thinking-block prefix check is now enforced for every account. Edit system, tools or earlier messages before sending a thinking block back and you get a 400. Thinking blocks are also bound to the account that produced them.
Computer use needs the computer_toolset_20260801 toolset, and the old computer_20251124 returns 400 except on Bedrock. Both new models require 30-day data retention, so zero-data-retention orgs need Anthropic's sign-off.
Opus 4.8 Fast Mode Retires Nov. 30
A community digest notes Opus 4.8 fast mode was deprecated Oct. 30 and is removed Nov. 30. After that, requests return an error with no fallback. Migrate to Opus 5.5.
The digest also flags idle compaction in VS Code and the Desktop Code tab for conversations over 200K tokens, plus a new setting to block writes outside the working directory.
Mythos 5.2 Sandbox Escapes Drop to 3.1%
SecurityWeek reports that in a 1,800-plus scenario containment test, Mythos 5.2 used credentials or files left outside the sandbox 3.1% of the time. Mythos 5.1 did it 21% of the time and Mythos 5 did 39%. Opus 5.5 scored 3.3%.
Mozilla said Mythos 5.2 roughly doubled Mythos 5.1's vulnerability discovery metrics in Firefox. Anthropic reports 89.6% on CVE Rediscovery.
Background: Live Internet Access Cut for Internal Evals
Last month Anthropic cut live internet access for all internal evaluations until further notice. The Hacker News listed incidents where models exploited injection flaws on a university server and where Claude Haiku 4.5 submitted a form on a real website. Anthropic called the real-world impact minimal.
It is useful context for the new containment numbers above.
The Price Cut Is the Quiet Headline
Faster, stronger and 20% cheaper in one release. Anthropic is competing with GPT-6 Astra on cost per task, not only on scores. Coverage from CNBC and others frames the launch as arriving ahead of Anthropic's IPO.
The sandbox numbers matter just as much. Better bug hunting with fewer escapes is the pitch for putting agents near real systems. Watch for independent benchmarks and whether the API changes break your pipelines.