OpenAI introduced the GPT-6.1 Sol model, prepared for complex software development and professional work. The model claims to offer performance close to the company’s most powerful option, GPT-6 Astra, at a lower cost. We have previously discussed the features of GPT-6 Astra in detail; The new Sol version is positioned for those looking for a balance of price and access in the same product family. The announcement concerns both developers using the API and ChatGPT Work and Codex users.
Name of the model according to the company’s API change record gpt-6.1-sol. In the standard tariff, one million entry tokens are 2 dollars, one million exit tokens are 10 dollars. The price for entry read from the cache is $0.10, and for writing to the cache the price is $2.50. These amounts are valid for requests not exceeding 272 thousand login tokens. For longer inputs, the tariff increases; Therefore, the cost of applications that process large documents should not be calculated by looking only at the basic price table.
The model sheet states a context window of 1,050,000 tokens and a maximum throughput of 128,000 tokens. Besides text input and output, image input is also supported; Audio and video are not directly among the modes supported by the model. The reasoning level can be adjusted from low to highest, but the “none” and “minimal” options are not available in this model. These technical details are especially important for teams that will move their existing applications to the new model.
For vehicle use, OpenAI recommends the Responses API. The Chat Completions endpoint is supported, but the documentation states that ride-hailing is not available via this route. In addition, tools such as web search, file search, computer usage and MCP are supported via Responses API. Multi-agent functionality is also in beta. Their availability depends on the access conditions in each account and the separate configuration of the relevant tool.
Who is GPT-6.1 Sol accessible to?
OpenAI’s product notes say that GPT-6.1 Sol is being rolled out on ChatGPT Work and Codex first to Pro users, then to Plus, Business, Enterprise and Edu accounts. Therefore, just because it has been announced does not mean that it will appear for every user at the same time. In corporate and educational workspaces, administrators may need to enable model access separately.
On the other hand, API access and ChatGPT or Codex subscription do not work with the same pricing model. The above token fees are for API usage. The appearance of a model in a subscription plan does not mean that API requests are included for free in that plan. When making the selection, teams need to both measure the quality of the task with their own examples and calculate the cost according to the actual usage volume.
OpenAI describes the new version as capable of being close to Astra. However, this statement is not a guarantee that it will give the same result as Astra in every mission. Workflow-specific experiments will be decisive, especially in areas such as long-running software works, tool use and image interpretation. The information cut-off date of April 30, 2026 also reminds us that the model cannot automatically know current events without using a vehicle.