OpenAI looks to be preparing GPT-6.1 Terra

Expand

Open original image ↗

OpenAI looks to have another Terra model on the way. We found gpt-6.1-terra in a merged pull request in Microsoft’s EF Core repository, where it is set as the default judge for a collection of coding evaluations. 1 4

The name appears in 18 evaluation configs. Seventeen were changed from GPT-5.6 Terra to GPT-6.1 Terra, while a new query-pipeline evaluation uses the same default. The model doing the tasks was also moved from GPT-5.6 Sol to GPT-6.1 Sol. This was part of PR #39217, merged on 16 October. 2

The tooling file makes the split easy to see. Sol handles the task; Terra is configured to judge the work. That config also sets medium reasoning effort and a ten-minute timeout for the judge. Those are evaluation settings, rather than published specifications for a new model.

A model name to go with the reporting

Earlier today, The Information’s Stephanie Palazzolo flagged a potential new Terra model from OpenAI in a public post about upcoming releases. We shared that report before finding the Microsoft reference. The code adds something useful: an exact GPT-6.1 Terra identifier, in a change that has already been merged. 5 6

EF Core is Microsoft’s .NET data-access project. Its evaluation harness tests repository instructions and coding skills, comparing runs with and without those customizations. A judge helps grade the resulting work and compare the runs. That is where Terra appears here. 3

Our guess is that a new Terra release is not far off. Microsoft has moved a whole set of evaluation defaults to the new generation, and the name is used consistently across the configs. That is a stronger sign of preparation, though the config alone cannot tell us whether Terra has run successfully or when OpenAI will release it.