GPT-6.1 Sol API Released: Migration Guide and Pricing
OpenAI launched GPT-6.1 Sol at DevDay on Sept 29, 2026, detailing pricing, reasoning effort updates, and migration steps.
Stock photo for illustration only, not from the actual event
- OpenAI officially launched the GPT-6.1 Sol model during DevDay on September 29, 2026.
- API requests require a POST call to https://api.openai.com/v1/responses with model gpt-6.1-sol.
- Support for none and minimal in reasoning.effort has been removed, requiring migration to low.
- Pricing remains $2 input and $10 output per million tokens, with cached input dropping to $0.10.
OpenAI officially rolled out the GPT-6.1 Sol model during its DevDay event on September 29, 2026, replacing older versions with legacy documentation now redirecting straight to this new release. This update brings refined performance metrics alongside critical structural parameter changes that developers must adopt to ensure seamless integration.
To interact with the API, developers need to send a POST request to https://api.openai.com/v1/responses, specifying the model identifier as gpt-6.1-sol along with a Bearer token for authorization. A standard implementation requires configuring key parameters:
- Set model name to gpt-6.1-sol
- Configure reasoning block with medium effort
- Provide test input text payload
Stock photo for illustration only, not from the actual event
One of the most critical breaking changes involves the reasoning.effort parameter. Previous values such as none and minimal are no longer accepted by GPT-6.1 Sol. Applications relying on those configurations must transition to low and thoroughly re-evaluate affected workflows, as submitting deprecated values will trigger an HTTP 400 error.
The transition from GPT-6 Sol to GPT-6.1 Sol highlights OpenAI's focus on balancing reasoning depth with operational expenditure. Mandating the low setting instead of allowing reasoning to be completely disabled ensures that models maintain logical rigor even during latency-sensitive operations. Developers should carefully audit automated pipeline workflows that previously depended on minimal effort settings.
Regarding pricing structures, GPT-6.1 Sol maintains input costs at $2 and output costs at $10 per million tokens, while cached input rates are reduced to $0.10. Furthermore, any prompt exceeding 272K input tokens is billed at double the standard input and cache rates, alongside 1.5x the output rate for the entire request. The knowledge cutoff date has also been updated from April 20 to April 30, 2026.
Before shifting production traffic entirely, OpenAI strongly recommends conducting comparative regression testing between both model IDs. Developers can build and execute test scenarios via CLI tools like Apidog to evaluate output accuracy, token usage, and overall expenditure before promoting updates to live environments.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment