Microsoft-Decision-1
Microsoft · Released 9 Oct 2026
A decision-scoring model for routing, classification, prioritization, verification, and workflow control. Microsoft said it was post-trained from Qwen3.5-9B for single-pass decision scoring and returns a calibrated probability for each option in a fixed set.
- Maker
- Microsoft
- Release date
- 9 Oct 2026
- First seen
- 9 Oct 2026
- Access
- Microsoft Foundry and OpenRouter.
- Pricing (as publicly stated)
- As stated in the 9 Oct 2026 announcement: $0.042 per million input tokens; output tokens free.
- Modalities
- text
- Question types
- yes/no, multiple-choice, rating, rubric grading
- Latency (as stated)
- Microsoft said the model's P50 latency was about 35 times faster than GPT-6 Sol in its benchmarking, and 2.5 times quicker than H2O-Lightning-4B v1.1, which it listed as the runner-up in that test.
- Official announcement
- https://commandline.microsoft.com/microsoft-decision-1-model-foundry/
Benchmark claims
These are claims as published by the maker or in the linked official post. We have not independently rerun every benchmark.
Microsoft said Microsoft-Decision-1 achieved the highest accuracy in its 36-benchmark comparison, spanning nearly 150,000 questions across benchmarks kept blind from training. It also reported that the model changed its decision on 1.3% of eight kinds of input perturbations on average, with zero flips when option descriptions were paraphrased or options were reversed or shuffled.
Names and marks belong to their owners. This page is an independent summary, not a product page from Microsoft.