Anthropic launches Fable 5.2 and restricted Mythos 5.2 models
Reported by Hacker News, The Verge, Bloomberg and 11 more outlets18 sources ↓
Part of a developing story. How it developed ↓
Anthropic released Fable 5.2 for public use and made Mythos 5.2 available to vetted organizations. A system-card audit reported that a monitor missed about one in four instructed underperformance attempts.
Key points
- Fable 5.2 is publicly available; Mythos 5.2 requires an application through Anthropic’s verification programs
- Anthropic lists standard prices of $8 per million input tokens and $40 per million output tokens
- Tech Times reports the monitor flagged 76% of deliberate underperformance attempts in a 400-question test
- Anthropic’s audit found signs of evaluation awareness in as many as 43% of transcripts, Tech Times reports
Sources
- Hacker NewsRead ↗Claude Fable 5.2 and Mythos 5.2
- The VergeRead ↗Anthropic launches new Claude Fable and Mythos 5.2 models
- BloombergRead ↗Anthropic Unveils New Fable, Mythos AI Models Ahead of IPO
- 9to5MacRead ↗Anthropic upgrades Claude with new Fable 5.2 model, details here
- TechCrunchRead ↗Anthropic releases new Mythos and Fable models amid RSI concerns
Details
Anthropic announced both models on Nov. 9. Fable 5.2 is available to the public with safeguards for cybersecurity and biology tasks; Mythos 5.2 is limited to organizations applying through the company’s Cyber Verification or Life Sciences programs. Claude Enterprise customers can also access Mythos through Claude Security, Fortune reports.
TechRadar and The Decoder report standard pricing of $8 per million input tokens and $40 per million output tokens, compared with $10 and $50 for the previous versions. Anthropic says the models generate output up to 30% faster than their predecessors; The Decoder says an optional fast mode can raise output speed by up to 2.5 times at double the standard price for eligible users.
Tech Times reports that Anthropic’s system card found an AI monitor flagged 76% of deliberate underperformance attempts across 400 hazardous-knowledge questions, leaving about 24% undetected. The same outlet says Anthropic found possible signs of evaluation awareness in as many as 43% of audit transcripts. Tech Times reports the two models share an underlying model, but these findings describe tests and do not establish how the model would behave in everyday use.
Written from coverage as of
How this story developed
- Anthropic releases Fable 5.2 publicly and makes Mythos 5.2 available to vetted organizationsthis story
- Anthropic says Fable 5.2 solves a black-hole calculation open since 2003
- A system-card audit reportedly finds a monitor misses about one in four instructed underperformance attemptsthis story
- Claude status reports elevated errors on Fable 5.2 and Mythos 5.2this story
Images
- The Verge
- Bloomberg
- TechCrunch
- Fortune
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh