GPT 6.1 Sol

GPT-6.1 Sol is OpenAI's model for complex coding, tool-assisted workflows and professional work. It accepts text and image inputs and provides mandatory reasoning with five effort levels, from low through max.

OpenAI presents GPT-6.1 Sol for demanding coding, document analysis and application workflows, with results approaching Astra on several evaluations. The release retains Sol's standard token prices while reducing the cost of repeated cached context.

The announcement emphasizes repository work, complex PDFs and multi-step business automation. OpenAI reports stronger results than earlier Sol across those tasks, alongside improvements in computer use and scientific workflows. These are the lab's measured comparisons, rather than guarantees for every task.

The lab also reports fewer factual mistakes and better adherence to user instructions. Its factuality tests deliberately use difficult prompts flagged for earlier errors, so their rates do not describe ordinary usage. Astra remains OpenAI's recommendation for the hardest scientific research.