Claude Fable 5.1 is live — the sticker price never moved, and the cache read fell 75%

Claude Fable 5.1 shipped today, alongside Claude Mythos 5.1. Our radar detected it at 17:52:22 UTC and confirmed it on the next sweep 60 seconds later.

The headline number is the one that did not move. Fable 5.1 lists at $10 per million input and $50 per million output — identical to Fable 5, exactly as the July 26 leak claimed. Read only the sticker and you would conclude nothing changed.

The price that did move is the cache read. Anthropic cut it 75%, from $1.00 to $0.25 per million tokens, and that single line does most of the work: the company puts the saving at roughly 25% on a typical workload and up to around 45% on complex coding and highly agentic tasks. For anyone running long agent loops — where the same context gets re-read on every turn — that is the entire economics of the model changing while the price card stays still.

What else is confirmed, from Anthropic's own post. Available today on the Claude API as claude-fable-5-1, and on AWS, Google Cloud and Microsoft Azure. A 1M-token context window with 128K max output, a June 2026 knowledge cutoff, and text, image and PDF in. Effort is selectable across low, medium, high, xhigh and max — and the defaults differ by surface: High in Claude Code, Medium in Claude Cowork and on Claude.ai. Anthropic also notes that at Low or Medium effort, 5.1 matches or beats Fable 5 at materially lower cost, which is a more useful claim than a benchmark bar because it is about the settings people actually run.

Mythos 5.1 shipped in the same announcement, carrying advanced biology capabilities, with enrollment for scientists said to be opening soon. It is the tier that spent June under export control, so its quiet return to a normal launch cycle is its own small milestone.

The market called this one. Polymarket's next-Mythos-class contract sat at 78% for September 30 when we wrote about it on August 20, moved to 90% four days later, and settled today. The August window it had already written down to 30% never happened, which is the part worth remembering: the leak that started this said August, and the money said September for weeks before September arrived.

Now the part about our own plumbing, because it is the reason this alert exists at all. The id that triggered detection was anthropic/claude-fable-5-1, and it came from models.dev's first-party Anthropic provider. OpenRouter still carries no 5.1 id at the time of writing. models.dev only became a launch source on August 14 — we added it after GLM-5.3 shipped on Z.ai's own blog and our correctly-armed entry never fired, because no source we swept carried the id. Without that fix, this launch would have been invisible to the alert for as long as OpenRouter took to list it.

That is two launches in three weeks where the deciding factor was not the pattern or the logic but whether anything we read carried the model at all.