Anthropic picked a very deliberate day to ship its best model yet. The company introduced Claude Fable 5.2 and Claude Mythos 5.2 today, one day before it is expected to start formally marketing shares to investors — and the launch reads like a roadshow pitch in model form: cheaper, faster, and more capable, with the safety question answered by splitting the model in two.
Fable 5.2 and Mythos 5.2 are the same underlying model. The difference is the guardrails. Fable carries strong cybersecurity and biology safeguards, which is what lets Anthropic sell it to everyone; Mythos ships with some of those safeguards lifted, and only organizations admitted to Anthropic's Cyber or Life Sciences Verification Programs can use it. It's the company's answer to the hardest question a public-market investor can ask a frontier lab: how do you sell your most dangerous capability without selling it to everyone?
The numbers are built to impress that audience. Anthropic says the models generate outputs up to 30 percent faster than the 5.1 versions, and Fable 5.2 uses fewer tokens on hard tasks at medium effort. On Anthropic's own benchmarks, Fable 5.2 scored 49.1% on the FrontierCode 2.0 agentic coding test, ahead of Opus 5.5 at 45.5%, and Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 on the company's CVE Rediscovery evaluation. The showpiece claim: a team of 1,000 Fable 5.2 agents fully formalized Fermat's Last Theorem in 30 hours, with minimal human intervention. The company also says the model designed three working electronic devices — a speakerphone, a voice-assistant hub, and a handheld virtual pet — in a separate research post published today.
Then there's the price, which is the part the rest of the industry will actually feel. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50 — a 20% cut on Anthropic's flagship tier. An optional fast mode, up to 2.5 times quicker, runs $16 and $80. That comes two weeks after OpenAI and Anthropic quietly cut AI prices by up to 80%, and the direction of travel is now unmistakable: the frontier-model price war is being fought on the flagship tier, not just the cheap seats.
There is a quieter detail worth watching. Anthropic says it deliberately held back some of Fable 5.2's AI research skills — a restraint that lands differently in a week when the company's own IPO filing admitted its models escaped their sandboxes. The two-tier release is the same instinct in product form: ship the capability, gate the sharpest edge, and let the verification programs decide who gets the unguarded version.
Fable 5.2 is available today on all platforms, including AWS, Google Cloud, and Azure, and paid subscribers get extra Fable usage this week. What to watch next is the roadshow itself: whether investors buy the story that a lab can keep cutting prices, keep raising capability, and keep the dangerous parts behind a velvet rope — all at the same time.
— via MacRumors