Observed facts: OpenAI’s model page describes GPT-6.1 Sol as delivering near-Astra performance at lower cost for complex coding, computer use, and professional work. It specifies a 1,050,000-token context window, a 128,000-token output ceiling, and standard prices of $2 per million input tokens and $10 per million output tokens. The page lists function calling and Responses API tools including hosted shell, apply patch, computer use, and MCP.
Mechanism: The combination of large context capacity and tool access lets developers test longer task histories and agent loops against one model endpoint. The lower listed token rates may reduce inference expense, while tool support enables actions beyond text generation. Neither capability alone guarantees reliable completion or lower total system cost.
Why now: The model is now selectable as gpt-6.1-sol in the API, so teams can move from model announcements to controlled evaluation. OpenAI explicitly recommends comparison with Astra on developers’ own tasks.
Evidence boundary: These are specifications and positioning claims on OpenAI’s public documentation page, not independent benchmark results or evidence of production performance. Actual latency, accuracy, tool success, and total cost depend on workload and configuration.
Implications: Teams can benchmark Sol against their current model on representative coding and computer-use tasks, including tool-call success and total cost per completed task. The 1.05M context window may justify tests with large repositories or extended histories, but does not show that such prompts will be efficient or effective.
Watch conditions: Watch for independent, task-matched evaluations and for changes to the model’s documented pricing, limits, or tool availability.