OpenAI has introduced GPT-6.1 Sol, adding another option to its GPT-6 model family. In a system-card addendum published on 29 September, the company says the model is designed to deliver capabilities close to GPT-6 Astra with a different balance of speed and cost.
That is a useful distinction for anyone choosing an AI model. The most powerful option is not automatically the right one for every task. A developer running many coding jobs, for example, may care as much about predictable costs and response time as about peak performance on the hardest request.
What OpenAI has published
The official API model page lists GPT-6.1 Sol at US$2 per million input tokens and US$10 per million output tokens at standard rates. It describes the model as suitable for complex coding, computer use and professional work, with a context window of 1,050,000 tokens. Those figures are platform specifications, not a promise that every user’s bill or result will look the same. Tool calls, processing modes and other options can change costs.
OpenAI says developers should use its Responses API for tool calling with the model. It also says users should compare Sol with Astra on their own tasks. That is sensible advice: benchmark scores and a product description cannot tell a team whether a model handles its particular files, language or workflow well.
The safety information deserves a read
The accompanying system-card addendum reports safety evaluations and says OpenAI is treating GPT-6.1 Sol as Critical in cybersecurity and High for biological and chemical capability under its Preparedness Framework. The company says it applies the safeguards used for Astra. These are OpenAI’s own assessments, and the document gives more detail than a launch headline can carry.
The practical takeaway is to test the new model on real work with appropriate review, especially before allowing it to take actions in connected systems. The launch expands choice; how much value it brings will be clearer once teams compare quality, speed and cost in the tasks they actually perform.
Sources & further reading
Prepared with AI assistance from linked reporting. The cover is an AI-generated editorial illustration. Spotted something we should correct?