Anthropic unveiled Claude Fable 5.2 and Claude Mythos 5.2 on Monday, November 9, Journal du Net reported. According to JDN, they are the same model with different guardrails: Fable 5.2 is for the general public, while Mythos 5.2 goes to vetted organizations through cybersecurity and life sciences verification programs and can handle advanced cyber and biology tasks that Fable will often refuse.
In benchmarks published by JDN, Fable 5.2 scored 49.1% on FrontierCode 2.0 versus 45.5% for Claude Opus 5.5, 82.1% on WANDR web research versus 72.3%, and 89.3% on OSWorld 2.1 computer use, ahead of GPT-6 Astra (79.5%) and Gemini 4 Carbon (81.5%). It trailed Gemini 4 Carbon on AutomationBench workflow automation, 51.4% to 53.9%. Anthropic says the model can generalize to some robotics situations it was not trained on.
Anthropic announced a price cut of about 20% versus Fable 5.1, to $8 per million input tokens and $40 per million output tokens, JDN reported. But Artificial Analysis measured Fable 5.2 at about $27 per task on its index, against about $10 for Opus 5.5 and about $6 for GPT-6 Astra. JDN also reported that Fable 5.2 can use up to 50% of weekly usage limits on standard Pro and Team plans, and that Artificial Analysis clocked it at about 98 tokens per second at max reasoning.
JDN framed the release as a bid to widen Anthropic's customer base ahead of its planned IPO.