OpenAI said on Nov. 17 in a thread on X1 that GPT-6.2 Astra and GPT-6.2 Sol are available in ChatGPT Work, Codex and its API.
OpenAI describes Astra as a frontier model for its users' hardest projects and Sol as an efficient model for everyday work. It also makes performance, cost and cybersecurity claims for the pair. This account rests on that thread, and Automatica Press did not review the openai.com page the thread links.
OpenAI's thread cites only its own evaluations and reports no independent test results. It says Astra "sets new highs on evals across professional work and scientific research at less than a third of the cost per task of GPT‑6 Astra." The thread gives no pricing or method for the cost comparison. It says Astra brings greater accuracy to engineering work and stronger visual reasoning for websites.
On cybersecurity, OpenAI calls Astra "our most capable model ever for cyber" and says it "jumps 22 points over GPT-6 Astra on ExploitBench Internal Port." The thread does not describe that benchmark or what the points measure. OpenAI said the models launched with tighter safeguards for higher-risk cybersecurity assistance. It said it is working to bring them to Daybreak, where approved defenders could use them with reduced refusals.
OpenAI describes Sol as able to run multistep agentic workflows, write clearly and be intuitive to collaborate with. It said many OpenAI employees use Sol as their go-to model for technical and non-technical work.
Access varies by plan. OpenAI said both models are in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. Pro 500 subscribers can use Ultrafast, an option the thread names but does not define, with either model. OpenAI said it is bringing Ultrafast to Pro 200 subscribers for the first time, starting with Sol. Developers can reach gpt-6.2-astra and gpt-6.2-sol in the API, with Ultrafast available.
GPT-6 Astra, the model OpenAI uses as its comparison, appeared in earlier coverage. Automatica Press reported on Nov. 142 that OpenAI had dissolved its robotics unit, betting that general models such as GPT-6 Astra can handle robotic control. That report does not address this release.
Ethan Mollick wrote on X3 that he used Astra before launch and found it "very uneven & weird." He said it was harder to steer than other models but could deliver well on an ambitious project, building a space elevator ride from scratch with Blender and Unity. His post is a single anecdote and not a systematic test.