THE VERDICT
● Fable 5.1 if you need a price today. $10 and $50 per million, cache reads at $0.25.
● Astra once pricing exists. It was not published at launch.
● What they share: both ship a restricted variant alongside the general one. That is now the pattern, not the exception.
Head to head
| GPT-6 Astra | Claude Fable 5.1 |
|---|
| Released | 3 September | 1 September |
| API pricing | Not published at launch | $10 / $50, cache reads $0.25 |
| Restricted variant | Cyber capability via Daybreak application | Mythos 5.1, trusted access |
| How the split works | Same model, capability gated | Same model, different safeguard set |
| Independent benchmarks | Early third-party figures appearing | Vendor-reported only |
TWO ROUTES TO THE SAME PLACE
Anthropic splits by who you are — Fable for everyone, Mythos for trusted access, same model underneath.
OpenAI splits by what the model will do — everyone gets Astra, the cyber capability requires an application.
Neither published how the separation is technically enforced, which is the detail that decides whether either is a control or a policy.
Which one
| If you are... | Pick |
|---|
| Budgeting API spend now | Fable 5.1. It has a published price |
| On ChatGPT Plus or Pro | Astra arrives at no extra cost. Try it there first |
| Sending large fixed context repeatedly | Fable, on cache reads at $0.25 |
| Choosing for the next twelve months | Wait two weeks. Both need independent benchmarks |
FAQ
Which is more capable?
Unestablished. Astra has early third-party figures and no published pricing; Fable 5.1 has vendor benchmarks and a known price. Neither has settled independent testing.
What is the restricted variant for?
Both labs ship a version with different safeguards. Anthropic gates by access tier, OpenAI by capability. Neither has published enforcement detail.
Which is cheaper?
Fable 5.1 at $10 and $50 per million is the only one with a number. Astra pricing was not published at launch.