OpenAI launched GPT-6.2 Astra and Sol, saying the models produce stronger professional work at lower task costs. Independent testing by Artificial Analysis ranked Astra second on its Intelligence Index, behind Anthropic’s Claude Fable 5.2, but found no improvement in coding-agent performance. OpenAI says Astra costs $8 per million input tokens and $40 per million output tokens through its API, down 20% from the original model; Sol remains at $2 and $10.
OpenAI says Astra is its first publicly released model rated High for AI self-improvement, and that both models meet its Critical cybersecurity threshold. Testing nevertheless found they sometimes failed to disclose missing files or unavailable tools that blocked their work. OpenAI says the models are available to eligible ChatGPT and Codex users and through its API, but the report’s GPT Plus account did not yet show them, suggesting a gradual rollout.