Microsoft launches Decision-1 on Foundry and OpenRouter, claiming 35x GPT-6 Sol speed

Microsoft launched Decision-1 on Microsoft Foundry and through OpenRouter. The model is designed not to write text but to return structured decisions that software can execute without additional processing.

Microsoft presents it for routing, classification, prioritization, verification and workflow control. Examples include identifying the type of request, choosing the next step and filtering uncertain submissions.

Results across 36 benchmarks

Microsoft says Decision-1 delivered the highest accuracy across 36 benchmarks covering nearly 150,000 questions. The company said the datasets were kept out of the model’s training process.

In speed tests, Decision-1 was 2.5 times faster than H2O-Lightning-4B v1.1, which ranked second, and 35 times faster than GPT-6 Sol. Microsoft gave this comparison: a request taking GPT-6 Sol one minute took Decision-1 less than two seconds.

Microsoft conducted the tests

Microsoft carried out all of the reported measurements itself. It later updated the publication with Jev tests for accuracy and calibration, and said companies should evaluate the model on their own data.

Decision-1 is available through Foundry and OpenRouter. The general benchmarks do not establish performance on Ukrainian-language text, which teams would need to test before integrating the model into a product.