GPT-6.1 Sol is a new option for teams building coding and tool-using workflows. OpenAI launched it on September 29, 2026, for Codex, ChatGPT Work, and the API. It reports improvements over GPT-6 Sol on software engineering, professional document work, computer use, and factuality—while positioning the model below GPT-6 Astra on price.
The immediate developer decision is not whether to switch every request. It is whether GPT-6.1 Sol can meet your quality bar on a specific workflow at a better cost per completed task.
GPT-6.1 Sol at a glance
| Item | OpenAI’s published detail |
|---|---|
| API model ID | gpt-6.1-sol |
| Standard input | $2 per million tokens |
| Cached input | $0.10 per million tokens |
| Output | $10 per million tokens |
| Product access | Codex and ChatGPT Work for eligible Plus, Pro, Business, Enterprise, and Edu users |
| Chat access | Not available in Chat at launch |
OpenAI says the new model is close to Astra on selected agentic coding and professional-work evaluations at about one-fifth of Astra’s standard input and output token rates. That is a company-reported comparison under particular settings, not a promise about every application.
What changed from GPT-6 Sol?
Coding and multi-step work
OpenAI reports a higher score on DeepSWE 1.1 for long-horizon software-engineering tasks. It also reports gains on AutomationBench, where agents use tools to complete business workflows. For a product team, these results make the model worth testing on bug fixes, constrained feature work, document processing, and supervised automation.
The useful measure is an accepted result: did the change pass tests, satisfy the brief, and stay within scope? A model can look inexpensive per token while becoming expensive if it needs repeated fixes or extra review.
Computer use and factuality
On OpenAI’s selected OSWorld 2.0 offline computer-use test, GPT-6.1 Sol performed better than GPT-6 Sol. Its difficult-prompt factuality evaluation also showed fewer answers containing errors at low reasoning effort. OpenAI explicitly notes that those error-inducing prompts are not representative of ordinary use. Keep verification steps for factual answers and approval boundaries for actions.
Pricing and caching
The cached-input rate of $0.10 per million tokens matters most when an application reliably reuses a long prompt prefix. Measure actual cache hits rather than assuming the discount applies to all input. OpenAI also announced an upcoming Ultrafast option for Codex, but that faster tier was not yet available at the announcement date.
Where it fits in a model stack
GPT-6.1 Sol is a candidate for routine-to-demanding coding, agent, and business tasks that need more capability than a budget model but do not always justify Astra. Keep a stronger-model or human-review path for ambiguous, high-impact, or repeatedly failing work. If you already route across GPT-6 Sol, Luna, and Astra, add Sol 6.1 as a measured alternative rather than replacing routes by name alone.
Test a small set of real tasks. Record output quality, tests passed, latency, total tokens, tool calls, retries, cache hits, human correction time, and the cost per accepted result. Re-run safety cases involving broken tools, restricted actions, and misleading source material. OpenAI’s system card addendum is useful context, but your application’s tools and permissions determine its actual risk.
Availability and migration cautions
OpenAI says GPT-6.1 Sol is available to eligible users in Codex and ChatGPT Work, not Chat, and through the API as gpt-6.1-sol. Check your account and current model documentation before changing a production integration. Test supported tools, context limits, reasoning settings, and billing in your own project.
For agentic software, continue to require human approval before production deployments, external messages, destructive data changes, or other consequential actions. A stronger model does not grant it broader authority.
How I can help you put it to work
I help teams evaluate model choices and build safe workflows through AI consulting and custom AI development. That includes API integration, evaluation sets, model routing, cost controls, and review gates. I can also bring those features into website development, mobile app development, and workflow automation.
Official sources
Frequently asked questions
What is GPT-6.1 Sol?
GPT-6.1 Sol is OpenAI's September 2026 upgrade to GPT-6 Sol, positioned for coding, professional work, and computer-use agents at lower token prices than GPT-6 Astra.
Where can I use GPT-6.1 Sol?
OpenAI says it is available in Codex and ChatGPT Work for Plus, Pro, Business, Enterprise, and Edu users, and through the OpenAI API as gpt-6.1-sol. It is not yet available in Chat.
How much does the GPT-6.1 Sol API cost?
OpenAI lists $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. Total task cost also depends on tool calls, retries, and review.
Should I replace GPT-6 Astra with GPT-6.1 Sol?
Not automatically. OpenAI positions Sol as a lower-cost option approaching Astra on selected evaluations, while Astra remains stronger for some demanding tasks. Compare both on representative work before changing production routing.