The building at 1515 Third Street in San Francisco, home to OpenAI’s headquarters when photographed in June 2025. Archival photograph: Coolcaesar / Wikimedia Commons, CC BY 4.0.
OpenAI’s new GPT-6.1 Sol model arrives with a clear proposition for people building AI workflows: greater capability without paying the company’s highest standard token rates. Announced September 29, it is an upgrade to GPT-6 Sol focused on coding, computer use and professional work.
OpenAI says the model approaches GPT-6 Astra on several of its evaluations while charging one-fifth of Astra’s standard input and output token prices. That is the developer’s reported comparison, not a finding from an independent test by this publication. The useful question for customers is how well those results translate to their own documents, code and repeated tasks.
Where it is available
The launch announcement makes an important product distinction. GPT-6.1 Sol is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. It was not yet available in Chat at launch. Developers can call the model through the API under the name gpt-6.1-sol.
The API changelog’s September 29 entry lists standard rates of $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens for prompts containing up to 272,000 input tokens. It also lists a separate cache-write rate and notes beta support for multi-agent delegation in the Responses API.
Token prices are only part of a workflow’s bill. Repeated attempts and long generated answers can change the economics, so a low headline rate is most informative when paired with a measured completion rate and total usage for a representative task.
A practical way to compare
For developers deciding whether to switch, OpenAI’s model-selection guidance recommends comparing models with the same inputs and keeping the lightest setting that meets the required quality. It also suggests considering how often a workflow runs and how quickly the result is needed. That gives customers a practical framework for evaluating a change.
A small comparison can ask practical questions: Did the model finish correctly? Did someone have to repair its output? Was the result delivered fast enough? Keeping the task and success criteria fixed makes the answer easier to interpret.
GPT-6.1 Sol gives existing users another option to test this week. The release’s real value will depend on dependable completed work per dollar, rather than the size of a benchmark improvement alone.
