Microsoft enters decision-model race with Decision-1

Microsoft now has its own AI decision model, Decision-1, aimed at fast, structured choices rather than open-ended chat. According to The Decoder, the model puts Microsoft into a category that already includes decision-focused systems from OpenAI, Cloudflare and Jev, as developers look for cheaper, faster ways to classify, evaluate and route work.

For creators, marketers, freelancers and small teams, the story is less about another chatbot and more about the plumbing behind AI workflows. A decision model is designed for moments when software needs a reliable structured answer, not a long paragraph.

What Decision-1 is built to do

Microsoft says Decision-1 handles classifications, evaluations and routing decisions. The company also says the model has the potential to guide and control agents in complex environments, according to The Decoder.

That makes it relevant for automation-heavy AI systems where a model may need to make many small calls, such as choosing a category, deciding whether an input passes an evaluation, or selecting the next route in a workflow. Those tasks are different from general-purpose writing or chat because the output is usually constrained and needs to be returned quickly.

In practical terms, this is the kind of model a developer might consider when the AI task is:

• assigning an item to a category;
• judging or scoring an input;
• routing a request to the next step;
• helping an agent decide what to do next.

The important point for non-technical buyers: decision models are not positioned as replacements for broad chat assistants. They are narrower systems for structured decisions that may run in the background of agents, apps and automation pipelines.

Microsoft’s claims on accuracy and speed

Decision-1 is based on Qwen3.5-9B, according to The Decoder. Microsoft says it was the most accurate model in its tests across 36 benchmarks covering nearly 150,000 questions.

The company also claims a speed advantage. According to Microsoft, Decision-1 is 2.5 times faster than the runner-up in its comparison, H2O-Lightning-4B. The Decoder notes that Cloudflare’s open-source Clef models, which are also based on Qwen, were not part of that comparison.

That omission matters for anyone evaluating vendors. Benchmark claims can be useful, but the exact set of models included in a test affects how broad the conclusion is. Teams considering Decision-1 should treat Microsoft’s figures as a starting point, then test the model against their own classification, evaluation or routing tasks.

This also fits a wider shift in AI tooling: not every job needs the largest general model. For repeated structured decisions, latency and cost can matter as much as conversational ability.

Availability and pricing

Decision-1 is available through Microsoft Foundry and OpenRouter, according to The Decoder. Input tokens are priced at $0.042 per million, while output tokens are free.

That pricing structure is notable because decision workloads can involve many short calls. If a system is making repeated classifications or routing decisions, input cost and latency become key factors. Free output tokens may also be relevant where the model returns short structured responses, though teams still need to calculate total usage based on their own workloads.

For small businesses and independent builders, the practical question is whether a dedicated decision model can reduce the need to send every small decision to a broader, more expensive model. The source does not state specific integrations beyond Microsoft Foundry and OpenRouter, so buyers should check their own stack before assuming it fits into an existing workflow.

A crowded category forming quickly

The Decoder places Decision-1 in the same emerging lane as decision models from OpenAI, Cloudflare and Jev. OpenAI has also moved into this area with a Decisions API for yes/no, category and rating tasks, which we covered in OpenAI adds Decisions API for yes/no, category and rating tasks.

Jev is described by The Decoder as the startup that kicked off the decision-model trend in mid-September. The outlet also argues that the idea spread quickly and was soon adapted using open small language models.

For users, the rapid movement suggests this category could become competitive fast. The near-term buying advice is straightforward: match the model to the task. If you need open-ended reasoning or long-form generation, a general chatbot model may still make sense. If you need repeated, structured choices inside an agent or automation flow, Decision-1 and similar models are worth watching.