Microsoft-Decision-1
A decision model that returns calibrated probabilities instead of text, for routing and classification in agent pipelines
Context window
Not published
Input / 1M tokens
$0.042
Output / 1M tokens
N/A (no generated text)
Provider
Microsoft
Priced per input token only, since it returns probabilities rather than generated text. Public preview in Microsoft Foundry (US and EU Datazone). · Data verified 2026-10-11
Microsoft-Decision-1 (October 9, 2026) is a decision-scoring model from Microsoft AI, post-trained from Alibaba's Qwen3.5-9B. Instead of generating text, it returns a calibrated probability for each of a fixed set of answer options, which makes it a fast, cheap router, classifier and judge for agent pipelines. Microsoft reports it is 35x faster than GPT-6 Sol at median latency.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Public preview in Microsoft Foundry; also listed on OpenRouter.
Strengths
- ✓Returns calibrated probabilities, not free text
- ✓Very low price ($0.042 per 1M input tokens)
- ✓Fast single-pass decisions (Microsoft-reported 35x faster than GPT-6 Sol at P50)
Best for developers who...
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| Microsoft 36-benchmark decision eval | Highest accuracy (Microsoft-reported) | About 150K held-out questions; no independent results yet |
Source: Microsoft Foundry blog
Compare Microsoft-Decision-1
Compare Microsoft-Decision-1 with any other model
Build a comparison →All model comparisons →