Anthropic's New Claude Models Are Cheaper and Less Likely to Refuse Your Questions
Two updated models promise stronger performance at lower cost, fewer frustrating refusals, and a privacy option that keeps business data off Anthropic's servers entirely.
Key points
- Anthropic launched Claude Fable 5.1 and Mythos 5.1 in mid-2025.
- Fable 5.1 costs roughly 25 percent less than its predecessor, up to 45 percent less for complex multi-step tasks.
- Fable 5.1 is available to all users now; Mythos 5.1 is limited to invited testers in Project Glasswing.
- A new enterprise privacy option will store customer data on the customer's own cloud servers, rolling out later this autumn.
Anthropic has spent months hearing the same complaints from business customers: the models cost too much, they retain data they shouldn't, and they refuse perfectly reasonable questions. Claude Fable 5.1 and Mythos 5.1 are the company's answer.
Our 1 September report covered the initial launch; the pricing and privacy details were confirmed there first.
What's actually different this time?
Fable 5.1 costs about 25 percent less on average than Fable 5. For complicated multi-step tasks where the model runs autonomously through many steps, the saving reaches around 45 percent, driven largely by cheaper handling of cached data, meaning information the model already processed and stored so it doesn't repeat the work.
Every CEO Dan Shipper, speaking to The Verge AI, called it "the strongest coding model we've used, but now it's fast, token-efficient, and crucially actually speaks like a normal person." Box CEO Aaron Levie told the same outlet that his company's automated assistant built on Fable 5.1 caught subtle details in data that Fable 5 missed on an identical test. A third tester, Lisan al Gaib, noted on benchmarks that Mythos 5.1 at its lower-effort setting matches the older Mythos model at full power, which matters because lighter settings cost less and respond faster.
Will it stop refusing perfectly ordinary questions?
Partially. Fable 5.1 has what Anthropic calls "more precise safeguards," making it less likely to block basic biology questions that tripped the previous model. Mythos 5.1 keeps the same restrictions as before, so that change is Fable-only.
On cybersecurity, Anthropic now allows Fable 5.1 to identify weaknesses in software code. Penetration testing, exploit generation and binary-based vulnerability scanning will still be redirected to a separate, more restricted Opus model.
The safeguard question carries some weight given what we reported on 21 August: a researcher found a simple conversation trick that pushed several Claude models past their built-in restrictions, and Anthropic hadn't pulled those models.
What does this mean for businesses worried about privacy?
Enterprises will soon have an option called Enterprise Frontier Safeguards, which stores conversation data on the customer's own cloud infrastructure rather than Anthropic's. Anthropic describes this as offering "complete privacy." The feature starts rolling out later this autumn.
Fable 5.1 is available now across all of Anthropic's platforms. Mythos 5.1 remains limited to Project Glasswing participants, a closed invite-only testing programme, as first reported by The Verge AI.
Common questions
Do I need to do anything to get the cheaper pricing?
No action required. Lower prices for Fable 5.1 apply automatically whether you access Claude through Anthropic's website or via API, the technical connection businesses use to plug Claude into their own software.
Can I join Project Glasswing to try Mythos 5.1?
Anthropichas not publicly opened sign-ups. The programme is currently invite-only, and Anthropic has not announced a wider release date.
