Microsoft’s new Decision-1 model costs $0.042 per million input tokens and nothing at all for output. That pricing tells you how fast this category has gone from novelty to commodity.
Decision models are built to make fast, structured calls instead of long-form answers. Microsoft now has one, and it’s late. OpenAI and Cloudflare already ship their own, and both follow Jev, the startup that started the trend.
What Decision-1 is supposed to do
Microsoft said Decision-1 handles classifications, evaluations and routing decisions. The company also said the model has “the potential” to “guide and control agents through complex environments.”
Note the hedge in that second claim. Microsoft is describing what the model might do with agents, not what it has shown it can do.
The benchmark claims leave out a key rival
Decision-1 is built on Qwen3.5-9B. Microsoft said it’s the most accurate model it tested across 36 benchmarks covering nearly 150,000 questions.
Microsoft also said Decision-1 runs 2.5 times faster than the second-place model, H2O-Lightning-4B. But the comparison doesn’t include Cloudflare’s open-source Clef models. Those are also built on Qwen, which makes them the closest competitor Microsoft could have tested. You’ll have to run that comparison yourself.
Where you can get it
Decision-1 is available through Microsoft Foundry and OpenRouter. With output tokens free, your bill depends almost entirely on how much context you send in.
A rough few weeks for the company that started it
Jev kicked off the decision model trend in mid-September. The idea took off quickly. Then larger players adapted it and moved past it using open small language models like the Qwen base under Microsoft’s model.
Jev has gone from launching a category to competing against Microsoft, OpenAI and Cloudflare in it in less than a month. And Microsoft is charging less than five cents per million input tokens.
