OpenAI priced the GPT-6 Astra API before most developers can use it

The New Stack reported on 4 September 2026 that OpenAI's GPT-6 Astra API remains partly closed to developers even after Wednesday's launch, and that OpenAI is compensating paid ChatGPT subscribers with usage-limit resets rather than fixing access for API customers so far.

What the article says is documented already

OpenAI has published Astra's API pricing and specifications, per The New Stack's reading of the API documentation: $10 per million input tokens, $50 per million output tokens, a 1.05 million-token context window, and support for up to 128,000 output tokens per response. The endpoint itself is still rolling out to most paying customers.

What OpenAI has said in public

Sam Altman posted on X, quoted verbatim by The New Stack: "first, sorry for the messy rollout. second, when we screw up, we try to make it right. third, we should be able to begin broad rollout to API customers and chatgpt subscribers in the near future. as usual we will start with pro subscribers."

OpenAI's engineering lead for Codex, Thibault Sottiaux, wrote in his own X post that the rollout will take "a few days" to complete, that OpenAI is bringing new systems and additional compute online as access expands, and that "many novel systems will operate at scale for the first time" during the rollout. OpenAI has not published a specific cause for the delay.

The banked-reset offer

Sottiaux said paid ChatGPT subscribers will receive one banked reset for every day they remain without Astra access, starting 3 September. A banked reset is a saved refill of the existing usage limit that a subscriber can spend later after hitting a limit, reusing a mechanism OpenAI has run on ChatGPT Work and Codex.

The New Stack notes that so far OpenAI has not announced anything similar for developers waiting on API access, and did not respond to the outlet's questions about the rollout, banked resets, or API timeline.

The other API surprise

Astra's API also introduces a new class of safety-triggered interruptions that look the same as a timeout in the API response, which The New Stack covered separately earlier this week. Any production workflow that treats API timeouts as retryable will treat these interruptions the same way by default, so a team that wraps Astra calls needs to inspect the failure details rather than only the timing.

The published prices matter more than the launch itself. Before broad access arrives, a team can already write down what one production workload would cost at $10 per million input and $50 per million output tokens with a million-token context, and decide whether Astra is worth its cost compared with Claude Opus 5 or Fable 5.1 for the same job. For anything beyond a prototype, the "few days" timeline and the unexplained rollout problems mean production work should not depend on Astra being callable this week, and the new safety-interruption class needs its own error handling before any retry logic runs on it.

Source

Original reporting: "Sorry for the messy rollout": OpenAI launched GPT-6 Astra, but developers are locked out by Amanda Caswell, The New Stack, 4 September 2026.