Key facts
- OpenAI said Astra scored 5.2 percentage points above GPT-6 Astra on Agents' Last Exam, a test of multistep professional tasks.
- OpenAI said Astra beat Anthropic's Claude Fable 5.2 by 5.5 points on that test and cost 5% less per task.
- Artificial Analysis put Astra at 62 on its Intelligence Index, behind Claude Fable 5.2 at 65, and Sol at 57.
- OpenAI said Astra is its first publicly released model rated High for AI self-improvement, and both models meet its Critical cybersecurity threshold.
- Sol keeps the $2 per million input and $10 per million output token rates of its predecessor, OpenAI said.
- OpenAI said the models are available through ChatGPT Work, Codex and the API, though a GPT Plus subscription did not show them.
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, saying both improve reasoning, coding, scientific research and professional workflows at lower processing cost. Astra is aimed at demanding assignments such as financial spreadsheets, legal research and interactive 3D design, and Sol at everyday work, the company said.
OpenAI cut GPT-6.2 Astra API token prices by 20% from GPT-6 Astra levels, to $8 per million input tokens and $40 per million output tokens. Sol keeps the $2 per million input and $10 per million output rates of its predecessor, the company said. Tokens are the chunks of text models process and generate.
OpenAI said Astra performed 5.2 percentage points better than GPT-6 Astra on Agents' Last Exam, a test of multistep professional tasks, and beat Claude Fable 5.2 by 5.5 points while costing 5% less per task. It said Astra scored nine percentage points above GPT-6 Astra on Terminal-Bench-Science 0.1. Those are company claims.
Independent testing by Artificial Analysis put Astra at 62 on its Intelligence Index, which combines nine tests of reasoning, scientific work, coding and knowledge into a score out of 100, behind Claude Fable 5.2 at 65. Sol scored 57, matching the original Astra at roughly 15% of its cost per evaluated task, according to the tester.
OpenAI said Astra is its first externally released model to reach the High threshold for AI self-improvement, a safety classification measuring how much a model can accelerate AI research. Both models meet OpenAI's Critical cybersecurity threshold, and the company said it strengthened monitoring and built a system called Control Room to halt suspicious model activity.
OpenAI said the models are available through ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users and through the API for developers. A GPT Plus subscription did not show the models, suggesting a slower rollout. OpenAI has not added GPT-6.2 to its Daybreak program for vetted cybersecurity researchers.
The launch follows Anthropic's Fable 5.2 release the previous week and Alibaba's Qwen 4 Max on November 16, which charges $1.50 per million input tokens and $4.50 per million output tokens.
