GPT-6 Astra release date and price: $10 per 1M input tokens

GPT-6 Astra came out on September 3, 2026, according to OpenAI's API changelog, and it costs $10 per million input tokens and $50 per million output tokens as of November 2, 2026. That is the GPT-6 Astra release date and price in two lines. It is the most expensive model in OpenAI's current API lineup, five times the price of GPT-6.1 Sol on input.

OpenAI's own changelog calls it "our most capable model, built for the hardest end-to-end work." The rest of this article sticks to what OpenAI's developer pages say. Where they don't say, we say so.

GPT-6 Astra release date

The changelog entry for GPT-6 Astra is dated September 3, 2026. It describes the model as built for reasoning, coding, computer use and research. The model ID is gpt-6-astra.

Two follow-up entries are dated September 29. One added Ultrafast mode in the Responses API, with US data residency only. The other added async tool calling, mid-turn steering over WebSockets and the ability to change reasoning effort mid-conversation.

The changelog entry does not list a price. The price comes from the pricing and model pages.

GPT-6 Astra price

OpenAI's pricing page lists these standard rates for prompts up to 272,000 input tokens, per million tokens, as of November 2, 2026:

Longer prompts cost more. Above 272,000 input tokens, Astra costs $20 per million input tokens and $75 per million output tokens, and the higher rate applies to the whole request. The model page adds a cache write price of $12.50, says Batch and Flex run at 50% of standard rates, and says Fast mode costs twice the applicable rate.

Context window and limits

The model page gives a 1,050,000-token context window, a maximum input of 922,000 tokens, a maximum output of 128,000 tokens and a knowledge cutoff of April 30, 2026. Astra works with Chat Completions, Responses and Batch. It does not support Realtime, fine-tuning or embeddings.

The changelog lists migration changes. There is no none reasoning effort, no custom temperature, top_p or logprobs, and tool calling requires the Responses API. Misalignment monitoring is active on supported Responses requests.

Rate limits depend on your usage tier. On the Build tier it is 5,000 requests and 1,000,000 tokens per minute. Tiers rise on their own as spend grows. The model page lists no waitlist or regional restriction.

Is GPT-6 Astra free?

Not in the API. It is billed per token at the rates above, and there is no free tier listed for it.

For ChatGPT, we could not confirm plan details. OpenAI's help center pages on Astra access and usage returned errors when we tried to read them, so this article does not state which ChatGPT plans include it or how many messages each gets. OpenAI's Codex models page says availability varies by plan, client, workspace settings and rollout, and notes that some paid plans leave out the Astra Extra High power preset. It points to the help center for limits.

GPT-6 Astra not showing in Codex

OpenAI's pages don't explain why the model may be missing, but they do show where it should appear. Based on the Codex models page as of November 2, 2026:

• In the CLI, start with codex -m gpt-6-astra, or run /model in a session.
• In the desktop app, pick Astra Light, Astra Medium or Astra Extra High from the Power menu, or choose the model under Advanced.
• If it is missing, plan, client, workspace settings and rollout are the documented variables.

We could not verify a minimum Codex CLI version for Astra. The page names version 0.159.0 for GPT-6.1 Terra, and gives no figure for Astra. Update the CLI first, since the latest releases in the changelog, 0.165.0 and 0.166.0, are newer than anything mentioned there.

What we don't know

OpenAI's pages we could read give no ChatGPT plan list for Astra and no explanation of any access limits. For newer, cheaper options, see our report on GPT-6.1 Terra, which costs $1 per million input tokens. Check the pricing page before you build, because rates change.