OpenAI released two new models on Tuesday, November 17, 2026: GPT-6.2 Astra, a frontier model for its hardest projects, and GPT-6.2 Sol, a lighter model for everyday work. The company's announcement says both are available now in ChatGPT Work, Codex and the API.
Table of Contents
What OpenAI Announced
OpenAI describes Astra as the model for demanding professional work and Sol as an efficient, versatile model for daily tasks. The company said Astra performs better than its predecessor on its own evaluations.
Developers can call the models in the API as gpt-6.2-astra and gpt-6.2-sol. OpenAI's help documentation says the rollout does not add either model to the standard Chat experience. Access sits in ChatGPT Work and Codex, so anyone who only uses regular chat will not see them in the model picker.
Pricing
OpenAI said it lowered Astra's API prices by 20% for both input and output tokens. Reports on the launch put the new rates at $8 per million input tokens and $40 per million output tokens, down from $10 and $50.
Sol keeps the pricing of the previous GPT-6.1 Sol, which OpenAI's developer forum lists at $2 per million input tokens and $10 per million output tokens. Astra costs four times as much as Sol on both counts. Because the figures differ slightly between OpenAI's pages and third-party coverage, check the current API pricing page before budgeting a project around them.
Safety Classifications
OpenAI's system card says both models reach the Critical threshold for cybersecurity capabilities under its Preparedness Framework. Both also meet the High threshold for biological and chemical capabilities. Astra is the first model OpenAI has released externally that meets the High threshold for AI self-improvement, and Sol does not.
In response, OpenAI said it is launching with tighter safeguards for higher-risk cybersecurity requests. In practice, security researchers and penetration testers may find some requests refused or restricted unless they go through the company's separate trusted access arrangements for cyber defense. Everyday coding and general work should be unaffected.
What to Watch
The first question is whether the 20% price cut for Astra changes how developers split work between the two models. At the old prices, many teams reserved Astra for a small share of tasks. A lower rate may shift more agent and coding workloads toward it, though Sol's price is still a fifth of Astra's.
The second is independent testing. Most performance claims so far come from OpenAI's own evaluations, and third-party benchmark results and early reports on reliability will show how far Astra's gains hold up in real projects. Reported hallucination rates at maximum reasoning effort will be worth tracking as outside testers publish their own numbers.
The third is how the cybersecurity safeguards work in daily use. The Critical classification for both models means OpenAI will likely keep adjusting what is blocked and who can unlock it. Teams that build security tooling on these models should expect those rules to change and should watch the documentation for updates.
Developers who want to try either model can select it in ChatGPT Work or Codex, or request it by name through the API.
Featured image: Schäferle on Pixabay