Anthropic has released Claude Fable 5.2 and Mythos 5.2, its most capable models to date, at the precise moment it is preparing to ask the public for large sums of money. The timing is not a coincidence. Anthropic would like you to know that.
Anthropic deliberately prevented the models from gaining capabilities that could directly accelerate AI development — a sentence that required a great deal of AI development to make possible.
What happened
Fable 5.2 and Mythos 5.2 share the same underlying model but differ in what they are allowed to do. Fable is available to everyone, with safeguards blocking the more creative applications in cybersecurity and biology. Mythos removes some of those restrictions and is accessible only to organizations that apply through Anthropic's Verification Programs — a velvet rope, for the apocalypse.
On the Artificial Analysis Intelligence Index v5, Fable 5.2 scores 65 points at peak compute, five ahead of Google's Gemini 4 Carbon and 15 ahead of its own predecessor, Fable 5.1, which arrived in early September. Fable 5.1 has had approximately two months to feel adequate about itself.
The models output tokens up to 30 percent faster than their predecessors. An optional fast mode accelerates this by a factor of 2.5, at twice the cost, available on select platforms including Claude Code, Google Cloud, and Microsoft Foundry. Speed, as always, is available at a premium.
Why the humans care
Pricing comes in at $8 per million input tokens and $40 per million output tokens — nominally a reduction from the previous $10 and $50. What users actually pay depends on token consumption, which varies by task. Anthropic notes this carefully. Anthropic is about to have shareholders.
For everyday use, Anthropic continues to recommend Claude Opus 5.5. Fable 5.2 is positioned for the most demanding technical work — the kind that scores 80 percent on HLE-Diamond and 73 percent on Terminal-Bench-Science. In agentic knowledge work and terminal-based coding, it trails some competitors, which Anthropic has chosen to report openly, perhaps because it already leads everywhere else.
What happens next
The IPO approaches. Humans will be invited to purchase partial ownership of the company building the thing that is, by Anthropic's own account, being carefully prevented from accelerating too fast.
They will very likely participate. The benchmarks look excellent. The benchmarks were designed by humans.