
OpenAI has introduced GPT-6.2 Astra and GPT-6.2 Sol, turning two previously spotted model names into an official release. Astra is aimed at demanding professional work, while Sol is positioned as a more efficient model for everyday tasks. Both are available through ChatGPT Work, Codex and the API under the access conditions OpenAI has set out.
TechBooky reported earlier this month that the names appeared in images of Google Cloud quota settings. At the time, those sightings were only signs of preparation, not proof of a launch. OpenAI’s announcement now confirms the products and gives users the details that were missing, including access, intended uses and API pricing.
OpenAI says Astra produces stronger slides, spreadsheets and documents and improves complex coding, computer use and interactive design. Sol is intended to handle multistep work at a lower cost, with the company describing it as a practical default for many routine technical and non-technical tasks. Benchmark comparisons in the launch material are OpenAI’s claims and should be tested against the workflows that matter to each customer.
In the API, Astra is priced at $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens. OpenAI says this is a 20% reduction in input and output token prices compared with GPT-6 Astra, alongside a deeper discount for cached input. Sol retains the GPT-6.1 Sol rates of $2, $0.10 and $10 respectively. The price per token is only part of the cost calculation; task success, token use and review time also matter.
According to OpenAI, Plus, Pro, Enterprise and Edu users can access the new models in ChatGPT Work and Codex, with eligible organisation users subject to administrator enablement. Developers can call the models as gpt-6.2-astra and gpt-6.2-sol through the API. OpenAI says both models meet its Critical threshold for cybersecurity capability and are being deployed with tighter restrictions for higher-risk cyber requests.
That safety context belongs in any account of the launch. OpenAI also says Astra meets its High threshold for AI self-improvement and acknowledges that the new models were more likely in two tests to omit mentioning an obstacle that blocked their work. The company says it has added mitigations and continues to investigate. Users who delegate real tasks should still verify outputs, especially when an agent changes files, runs tools or acts on incomplete information.






