Guides · Models
GPT-6 Astra: price, context window and Codex use
| Developer | OpenAI |
|---|---|
| API release | 3 September 2026 |
| Model ID | gpt-6-astra |
| Context window (API) | 1,050,000 tokens (922,000 max input, 128,000 max output) |
| API price per 1M tokens | $10 input / $1 cached input / $50 output |
| Reasoning effort (API) | low, medium, high, xhigh, max (no none) |
| ChatGPT | Plus, Pro, Business and Enterprise (Enterprise admins must enable it) |
| Codex | Desktop app, web, CLI and IDE extension; 5 to 45 local messages per five hours on Plus (estimate) |
| Also on | Microsoft Azure and Amazon Bedrock, per OpenAI |
GPT-6 Astra is our most capable model for the most demanding work.
What OpenAI says it's for
OpenAI's GPT-6 guide calls Astra its most intelligent model yet, with state-of-the-art performance in computer use, browsing, software engineering, science and professional work. It says Astra is more likely than earlier models to ask a focused question when the answer could change the outcome, and that in several evaluations it used substantially fewer output tokens. In its announcement, OpenAI reports 57.9% on Terminal-Bench 4.0, ahead of the models it compared. Those are OpenAI's own results.
Safety checks you may notice
OpenAI says Astra reaches the critical cybersecurity threshold in its Preparedness Framework, so it ships with extra safeguards. Astra refuses more advanced offensive security work, such as writing proof-of-concept exploits. Misalignment monitoring runs asynchronously during agent work: in ChatGPT and Codex a flagged task may pause and ask you to verify the action, while in the API it stops.
Pricing details
- $10 input, $1 cached input, $12.50 cache writes and $50 output per million tokens.
- Prompts over 272K input tokens are billed at 2x the input and cache rates and 1.5x output for the whole request.
- Batch and Flex cost 50% of Standard. Fast mode costs 2x. Ultrafast (service_tier "ultrafast") is available in the Responses API with global processing and US data residency only.
Using Astra in Codex
- OpenAI's Codex docs suggest Astra for the hardest end-to-end work and starting at Light reasoning, which is low in config.toml.
- Plus and Standard Business: an estimated 5 to 45 local messages per five-hour window. Pro currently has no five-hour limit.
- Credit rate: 250 credits per million input tokens, 25 per million cached input and 1,250 per million output.
- Ultrafast on Pro $500 generates tokens up to 8x faster than Astra in Standard mode, and uses included usage at 8x the Standard rate.
- Select it with codex --model gpt-6-astra or /model in the CLI.
Independent benchmarks
Artificial Analysis lists GPT-6 Astra (max effort) at 53 on its Intelligence Index v4.3.2, at $3.26 per Intelligence Index task, as of 13 October 2026. In its 12 October comparison, Astra used about 27k output tokens per task, the fewest of the models it listed.
Questions
- GPT-6 Astra or GPT-6.1 Sol?
- OpenAI's advice is to use Astra for the hardest end-to-end work and GPT-6.1 Sol when cost matters, comparing both on the same task. Per token, Astra costs five times as much ($10 / $50 against $2 / $10).
- Can I use GPT-6 Astra on ChatGPT Plus?
- Yes. OpenAI's announcement lists Plus, Pro, Business and Enterprise, and its Codex pricing page gives Astra usage estimates for Plus. On Enterprise, access is off by default until an admin enables it.
- Why did Codex pause my Astra task?
- OpenAI runs misalignment monitoring and extra cyber safeguards for Astra. When they flag something in ChatGPT or Codex, you may be asked to verify the action before the task continues.
Sources
- GPT-6 Astra model page · OpenAI API Docs
- OpenAI API changelog · OpenAI API Docs
- Using GPT-6 · OpenAI API Docs
- GPT-6 Astra announcement · OpenAI
- Codex models · OpenAI · ChatGPT Learn
- Codex pricing and usage limits · OpenAI · ChatGPT Learn
- Speed: Fast and Ultrafast modes · OpenAI · ChatGPT Learn
- Reasoning models: reasoning effort · OpenAI API Docs
- Artificial Analysis: GPT-6 Astra model page · Artificial Analysis
- Artificial Analysis on Muse Wonder: Intelligence Index, cost per task, Coding Agents Index · X · @ArtificialAnlys ·
Last checked . Facts change; the sources above are authoritative. See how we verify or report a correction.
