OpenAI announced GPT-6.1 Sol on September 29, 2026, describing it as an upgrade to GPT-6 Sol for agentic coding, computer use and professional work. Its API documentation lists a 1,050,000-token context window, a maximum output of 128,000 tokens, and standard US rates of $2 per million input tokens and $10 per million output tokens.
OpenAI positions GPT-6.1 Sol as a GPT-6 Sol upgrade
OpenAI’s model announcement presents GPT-6.1 Sol as an update to GPT-6 Sol, not as a model for a single task. Its stated focus spans coding agents, computer use and professional work. The API model ID is gpt-6.1-sol.
GPT-6 Astra appears separately in OpenAI’s benchmark comparisons. For context on that model, see our earlier coverage of GPT-6 Astra.
What OpenAI’s named tests report
OpenAI reports different results on two evaluations: one compares Sol with GPT-6 Astra on software-engineering tasks, while the other compares it with Claude Opus 5.5 on multi-step business workflows.
| Evaluation and setting | OpenAI-reported result | Comparison and cost |
| DeepSWE v1.1 | GPT-6.1 Sol matched GPT-6 Astra | About one-fifth of GPT-6 Astra’s cost on the evaluation |
| AutomationBench 1.0.6, medium reasoning effort | 2.2 percentage points above Claude Opus 5.5 | About one-third of Claude Opus 5.5’s cost per task |
DeepSWE v1.1 evaluates long-horizon software tasks in real codebases; AutomationBench covers multi-step business workflows. OpenAI also reports that Sol scored 4.8 percentage points above GPT-6 Sol on AutomationBench at the same reasoning setting. These are company-reported results for named evaluations, not a guarantee of similar performance on every workload.
OpenAI’s safety evaluations
OpenAI classifies GPT-6.1 Sol as Critical for cybersecurity, High for biological and chemical capabilities, and below the High threshold for AI self-improvement under its Preparedness Framework. The company says Sol receives the same safeguards stack as GPT-6 Astra.
OpenAI’s system-card addendum also reports a 1.50% coding-misrepresentation rate for Sol, compared with 0.51% for Astra, in a deliberately challenging evaluation. In a separate test of unwanted persistence, Sol showed persistence in 23.5% of rollouts, versus 17.4% for Astra. OpenAI says these tests were designed to elicit failures and do not represent typical-use rates; the persistence test chiefly involved low-stakes restrictions and did not include system-level controls intended to prevent circumvention.
Access, API limits and US rates
OpenAI announced API access and access through ChatGPT Work and Codex for eligible paid plans. Its September 29 release-note entry described a rollout beginning with Pro and expanding to Plus, Business, Enterprise and Edu. At the time of the announcement, OpenAI said Sol was not yet available in standard Chat mode.
The API documentation gives these standard US rates per million tokens:
- Input: $2
- Cached input: $0.10
- Cache writes: $2.50
- Output: $10
Requests with more than 272,000 input tokens are charged at twice the input and cache rates and 1.5 times the output rate for the entire request. The model supports text and image input; audio and video input are unsupported. For tool calling, OpenAI directs developers to the Responses API. Chat Completions is supported without tool calling.
Dots is a separate agent feature
OpenAI introduced Dots at DevDay as always-on agents with their own cloud computer. They return results for the user to review, and the release notes say they are powered by GPT-6 Astra—not GPT-6.1 Sol. Pro and Business Premium users are eligible in supported markets; Enterprise, Edu and Healthcare users can try a beta when an administrator enables it. Enterprise access is off by default.
DevDay also included other Codex and API announcements, covered in our earlier OpenAI DevDay roundup.
