On Tuesday, November 18, 2026, OpenAI introduced the OpenAI GPT-6.2 Astra and Sol models. Astra targets demanding professional work, research and programming, while Sol targets everyday tasks. The API price for the Astra model has fallen, while Sol holds its price.
OpenAI GPT-6.2 Astra and Sol: what they bring
According to OpenAI, Astra is the company's most capable model for complex professional tasks, scientific research and engineering. Sol handles routine work ranging from writing texts to multi-step assignments. Both models are available to users in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers, and via the API they carry the names gpt-6.2-astra and gpt-6.2-sol.
API prices and test results
GPT-6.2 Astra costs 8 dollars per million input tokens and 40 dollars per million output tokens. That is a fifth less than GPT-6 Astra, which previously cost 10 and 50 dollars. Cached inputs come to 0.40 dollars per million tokens, 60 percent cheaper. Sol remains at 2 dollars per million input tokens and 10 dollars per million output tokens.
In the Agents' Last Exam test, Astra scored 5.2 percentage points more than the original GPT-6 Astra, at less than a third of the cost per task. It beat Anthropic's Claude Fable 5.2 model by 5.5 points at roughly five percent of the cost. On Terminal Bench Science 0.1 it improved by nine percentage points. According to OpenAI, Astra is the company's first publicly released model to meet the high level for AI self-improvement under the Preparedness Framework.
Testers' reservations and independent measurement
Early testers described recurring problems: looping without benefit, ending tasks prematurely and less comprehensible answers. One project ended up worse than before after days of code revisions. Thibault Sottiaux of OpenAI said on X that the company made substantial changes in the last three to four days before launch, which he said some reviewers did not register.
OpenAI speaks of record results. However, according to KuCoin, independent testing by Artificial Analysis assigned the Astra model an intelligence index of 62, slightly below the Claude Fable 5.2 model with a value of 65.