OpenAI's latest upgrade has a twist: the cheaper model is now almost as good as the expensive one.
The company has released GPT-6.2 Astra and GPT-6.2 Sol, new versions of the two models at the heart of its GPT-6 lineup.
Both now run in ChatGPT Work and Codex, OpenAI's coding tool.
OpenAI says both models can complete demanding tasks using fewer output tokens, the units of text an AI model generates, so users' allowances go further.
The everyday option closes in
Sol is the cheaper model meant for everyday use.
On Cognition's FrontierCode 2.0 benchmark, which tests whether code changes are good enough for a project's maintainers to accept, it scored within half a point of Astra.
It did so at about a fifth of the cost, with both models running at maximum reasoning effort.
Astra is still OpenAI's recommended model for the hardest work.
Given a screenshot of one slide, it can apply that slide's layout, colours and font sizes across a whole presentation.
OpenAI says it also makes cleaner, better-tested changes to existing code in Codex, with early testers reporting "cleaner code style and more focused tests, with fewer rounds of revision".
A fast-moving lineup
The release continues a staged rollout that began with GPT-6 Astra in September.
GPT-6 Sol followed, then GPT-6.1 Sol at OpenAI's DevDay event, which the company said matched Astra on agentic coding and professional work at a fifth of Astra's standard token price.
The timing is pointed: on the same day, Microsoft claimed that its new MAI Code 2 model reaches 98% of GPT-6.1 Sol's performance at 40% of the price.
Read more about OpenAI
- 17/11/26Microsoft bets "nearly as good" can beat OpenAI and Anthropic on price
- 26/10/26OpenAI adds Terra to its GPT-6.1 model lineup
- 04/09/26OpenAI launches GPT-6 Astra: What you need to know
Faster, at a price
Both models are available now in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and through OpenAI's API.
The $200-a-month Pro plan gets Ultrafast mode for Sol for the first time.
The $500-a-month Pro tier, along with eligible Enterprise and Edu workspaces, gets Ultrafast on both models once an admin switches it on.
Ultrafast generates tokens up to 4.5 times faster than standard, but eats through usage allowances more quickly.
GPT-6.2 Astra's API price is 20% lower than GPT-6 Astra's, while Sol keeps its GPT-6.1 pricing.