OpenAI Launches GPT-6.2 Astra and Sol with Enhanced Performance

On November 17, OpenAI introduced GPT-6.2 Astra and GPT-6.2 Sol – updated AI models for professional and everyday tasks. The company claims improvements in working with documents, financial spreadsheets, presentations, code, and interactive design.
GPT-6.2 Astra is designed for complex projects requiring sequential execution of multiple work stages. GPT-6.2 Sol is aimed at a broader range of tasks, including text writing, programming, and using digital tools.
How much do GPT-6.2 Astra and Sol cost
OpenAI has reduced the standard cost of using GPT-6.2 Astra via API by 20% compared to GPT-6 Astra. Now processing 1 million input tokens costs $8, and generating the same volume of output tokens costs $40.
The cost of GPT-6.2 Sol remains at the previous version's level: $2 for 1 million input tokens and $10 for 1 million outputs.
However, savings may depend not only on rates. According to OpenAI, the new models can perform some tasks using fewer tokens. This is especially important for multi-stage processes where AI sequentially works with documents, programs, and external tools.
The company also claims that Astra more accurately follows presentation templates and table formatting requirements, creates more consistent documents, and uses template phrases less frequently.
What independent tests showed
According to the Artificial Analysis Intelligence Index, GPT-6.2 Astra scored 62 points. This is five points higher than GPT-6 Astra's result, but below the score of Claude Fable 5.2 from Anthropic, which scored 65 points.
GPT-6.2 Sol scored 57 points, approaching the overall result of the original GPT-6 Astra. Meanwhile, according to Artificial Analysis, the cost of completing a test task for Sol is about 15% of the corresponding figure for the previous Astra.
However, improvements were uneven. In a separate Coding Agent Index, which evaluates AI agents' work on programming tasks, GPT-6.2 Astra did not surpass its predecessor. The result for GPT-6.2 Sol was three points lower than GPT-6.1 Sol.
More noticeable achievements were recorded in interactive 3D design. In the AA-3D test, the Astra model took first place with a score of 1318 Elo, surpassing Claude Fable 5.2 with 1271 Elo.
Thus, the update improved overall performance and efficiency in a number of professional tasks, although it did not ensure growth in all testing areas.
Why OpenAI strengthened security control
Along with the launch of the new models, OpenAI announced additional security measures.
GPT-6.2 Astra became the company's first model released for external users to reach the High level in the AI Self-Improvement category. This classification reflects the system's ability to accelerate AI research and development.
Both models also reached the Critical level in cybersecurity capabilities. In this regard, OpenAI has tightened restrictions for potentially dangerous requests and expanded monitoring of the models' operation.
One of the new tools is the Control Room system, designed to monitor research processes involving AI and stop the model's operation if intervention is needed.
However, OpenAI's testing revealed persistent shortcomings. In some scenarios, the models did not inform users of obstacles preventing task completion, such as a missing file or an unavailable tool. The company announced additional measures to reduce such cases.
Where the new models are available
GPT-6.2 Astra and GPT-6.2 Sol became available in ChatGPT Work and Codex to users of supported Plus, Pro, Business, Enterprise, and Edu plans. Corporate and educational workspaces may require administrator permission.
Developers can use the models through the OpenAI API.
However, Astra and Sol do not necessarily appear in the standard model list of a regular ChatGPT chat: according to OpenAI documentation, a Work mode is provided for their selection.
The new models expand OpenAI's capabilities in professional tasks and allow for cost reduction in certain workflows. However, independent results show that their advantages depend on the specific usage scenario.
This material is prepared solely for informational purposes and does not constitute financial advice or a recommendation.




