GPT-6 Astra

OpenAI's new top model, released 2026-09-03 per its own API changelog ("Released GPT-6 Astra, our most capable model, built for the hardest end-to-end work", tagged v1/responses and v1/chat/completions). Text+image input, text output; 1,050,000-token context, 128,000 max output, Apr 30 2026 knowledge cutoff. reasoning.effort supports low/medium/high/xhigh/max - the 'none' level is NOT supported, and neither are custom temperature/top_p or logprobs; tool calling requires the Responses API. Standard tier per 1M: $10.00 in / $1.00 cached in / $12.50 cache write / $50.00 out. Batch and Flex are 50% of Standard ($5.00/$0.50/$6.25/$25.00); Fast mode is 2x ($20.00/$2.00/$25.00/$100.00) and is unavailable with EU data residency. Prompts >272K input tokens are billed at 2x input and cache rates and 1.5x output for the full request. Rate limits are published for usage tiers 1-5 (Free: not supported). STATUS IS A JUDGEMENT CALL, recorded here rather than hidden, and it MOVED on 2026-09-05. From 2026-09-04 this row read 'preview' - not OpenAI's word, which it has never used for this model, but this catalog's way of not telling a consumer they could call it that day, because the model page then carried "GPT-6 Astra is rolling out today for enterprises in our Trusted Access Program, with access through API and our Plus, Pro, Business and Enterprise plans coming in the coming days." That sentence is now gone from both surfaces that carried it: the model page (fetched first-hand 2026-09-05) and the pricing page (check-pricing-prose reported it as a removed line the same day). The 2026-09-04 row pre-registered the trigger - "move to ga when that rollout banner drops or OpenAI states GA" - and the first half is met, so it is honoured here rather than re-argued. Be clear about how strong that evidence is: the trigger is a REMOVAL, not an assertion, and OpenAI has still published no explicit "generally available" for this model - its changelog does not repeat the caveat and does not lift it either. What corroborates the flip is the page's positive content, which now reads like every other GA row: full endpoint and tool support, published snapshots and aliases, Standard-tier pricing on the general pricing page, and no access qualifier anywhere on it. Changing this catalog's own inference is a correction, so it files no changelog event (LEARNINGS #48).

Source: OpenAI official documentation ↗

Explore more

GPT-6 Astra — questions & answers