GPT-6.2 Astra and Sol arrive in Codex and the API
- Source
- x.com
- Date
Upgrades, people. Upgrades. The wait is over, you can now use GPT-6.2 Astra and GPT-6.2 Sol in Codex and the API.
Context
OpenAI's developer account announced that GPT-6.2 Astra and GPT-6.2 Sol can now be used in Codex, OpenAI's coding , and through its API. The post itself says nothing more, but OpenAI's API documentation shows what changed for developers. It describes Astra as the most capable model for the hardest projects and Sol as the everyday model for coding and running agents at scale. The GPT-6.2 Astra page lists $8 per million input and $40 per million output tokens, below the $10 and $50 on the GPT-6 Astra page. The GPT-6.2 Sol page lists $2 and $10, the same as GPT-6.1 Sol. Both new models take text and images, answer in text, and accept up to 1,050,000 tokens of .
Ultrafast, the premium speed tier OpenAI introduced in September with claims of up to 8 times faster token generation in Codex, is priced differently for the new models. Both GPT-6.2 pages charge 3 times the standard rate for Ultrafast through February 28, 2027, and 6 times from March 1, 2027, while the GPT-6.1 Sol page lists Ultrafast at 6 times standard. The cheaper speed tier is a launch-period rate with a stated end date.
One change affects existing integrations. According to OpenAI's guide to using GPT-6.2, starting with these models the API drops the model's saved reasoning from earlier turns when it detects edits to earlier parts of the conversation, a switch to a lesser model, or a change of end user or organization. OpenAI warns that answer quality can then decline and the model may spend more tokens repeating earlier work. Apps that rewrite their instructions or earlier tool calls between requests are the ones to review. OpenAI recommends leaving the start of the conversation unchanged, which also keeps that input eligible for the discounted cached rate.
The video attached to the post is a demonstration, available here only as an automatic transcript of its audio, with the visuals not analyzed. The presenter, filmed in front of a green screen, says GPT-6.2 Astra in Codex rebuilt the studio set in the 3D program Blender from two frames of the video, iterating until the render matched the footage, and then turned it into a scene that could be explored like a video game. That is OpenAI's own showcase, and a transcript cannot show how close the match was or how much guidance the model received. The documentation establishes per-token prices, not task costs. Whether the models finish work for less depends on how many tokens they use. OpenAI says GPT-6.2 uses fewer, but none of the material read here measures it.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…







