OpenAI released GPT-6.1 Sol on September 29, 2026. It is aimed at complex coding, computer use, and professional work. OpenAI describes its performance as close to GPT-6 Astra, its most capable model, but at a lower cost. That is OpenAI's assessment, not an independent benchmark from Core Infinite.
The release gives teams another option for work they do often, such as updating software, reviewing documents, or working across apps. The main question is whether it can deliver the quality they need at a cost they can sustain.
What the model can handle
GPT-6.1 Sol accepts text and images and produces text. Its listed context window is 1.05 million tokens, with up to 128,000 output tokens. In simple terms, it can work with a large amount of material in one request. That does not mean every detail will be understood correctly, so important results still need review.
Developers can connect tools through the Responses API. Supported tools include web search, file search, code execution, and computer use. Chat Completions is also supported, but without tool calling for this model.
A lower price for repeated work
At launch, standard API pricing is $2 per million input tokens and $10 per million output tokens for prompts with up to 272,000 input tokens. Cached input costs $0.10 per million tokens. These are API rates, not a ChatGPT subscription price. Longer prompts, processing options, and tool use can change the bill.
Our takeaway is to compare the cost of a finished task, not just the price of tokens. A cheaper request is less useful if it needs several retries or leaves someone with hours of cleanup.
Several agents can share a task
GPT-6.1 Sol also supports OpenAI's multi-agent beta in the Responses API. A main agent can hand separate parts of a job to other agents, then bring their findings together. OpenAI lists codebase research, document comparisons, and independent implementation tasks as possible uses.
This can help when the work can happen in parallel. It can also use more tokens. A small task, or one where every step depends on the last, may be better suited to a single agent.
Where it is available
OpenAI's September 29 release notes list GPT-6.1 Sol in Codex and ChatGPT Work. Access depends on the plan, app, and workspace settings. Developers can select it in the API using the model name gpt-6.1-sol.
For product teams, our advice is to try it on a few real tasks before making it the default. Check the result, the time taken, and the total cost. Keep clear approval steps for actions that change files, accounts, or published content. A stronger model should make the work easier to review, not remove the review altogether.