OpenAI has released GPT-6.1 Sol, a reasoning model designed for agentic coding, computer use, complex documents, and multi-step professional workflows.
OpenAI says the model approaches GPT-6 Astra on several evaluations while charging one-fifth of Astra’s standard input and output token prices. For standard API requests with up to 272,000 input tokens, GPT-6.1 Sol costs $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens.
The company reports meaningful gains over GPT-6 Sol. On DeepSWE 1.1, GPT-6.1 Sol improved the previous model’s best result by 6.4 percentage points. On OSWorld 2.0 computer-use tasks, it came within 2.1 points of Astra at roughly one-seventh of the cost per task. These are vendor-reported results and should still be validated against each product’s own workloads.
Developers can access the model as gpt-6.1-sol through the OpenAI API. It has a 1.05-million-token context window, supports OpenAI’s agent tools, and includes beta multi-agent delegation through the Responses API. It is also available to eligible users in Codex and ChatGPT Work, though it is not yet available in standard Chat.
Why it matters
The important change is economic: near-frontier capability is becoming practical for agents that run frequently, maintain large cached contexts, or execute long workflows. Builders can reserve Astra for the hardest edge cases while routing more coding, document, and computer-use work through Sol.
That extends the capability-and-cost ladder introduced with GPT-6 Sol and Luna. Teams can now route more demanding steps to a model positioned close to Astra, while using Luna for focused, high-volume work and reserving Astra for the comparatively small number of tasks that need maximum capability.
This could make production agents substantially cheaper without dropping all the way to a lightweight model—but teams should compare completion quality, total task cost, latency, and safety behavior on their own evaluations before switching.