OpenAI released GPT-6.1 Sol on 29 September. It is actually available in ChatGPT Work, Codex and the API; ordinary Chat is not yet among the supported surfaces. The company positions it as a stronger successor to GPT-6 Sol, aimed particularly at coding, complex documents and agents that must complete several connected steps.

In OpenAI's tests, it approaches the more expensive Astra on selected tasks. The company lists API prices of $2 per million input tokens and $10 per million output tokens. The one-fifth-price comparison concerns standard token rates against Astra, not necessarily the cost of every completed task. On a set of difficult prompts where users had previously flagged factual errors, the share of erroneous answers at low reasoning effort fell from 11.4% to 7.7%. That set is not representative of normal use, and the results cannot yet be generalized to every business.

If independent use confirms the improvement, the model could make document review, routine software fixes or assistants linking several tools more economical. In those workflows, a benchmark score alone is not decisive: total task cost, human corrections and the ability to stop before an unauthorized action matter just as much.

Because the API is already available, limited pilots could reasonably be evaluated within weeks to three months. Wider use requires testing on an organization's own data, step-specific permissions, auditability and safeguards for sensitive information; there is no defensible fixed date for that. The pilot window is an editorial estimate, not an OpenAI promise. The image is an abstract computing illustration, not a product screenshot.