Microsoft unveils Microsoft-Decision-1, a fast decision-scoring model trained on Qwen3.5-9B, and says it will soon rebase it on MAI, OpenAI, and other models
Decision models are quickly emerging as an important new category in AI. Unlike LLMs, which are designed to generate text …
Command LineAchint Srivastava
Context & Ripple Effects
Microsoft has been building a broader in-house model portfolio: its MAI-Thinking-1 reasoning model was presented as trained without third-party distillation, while its June releases paired reasoning and efficient coding models for business use and lower operating costs. Decision-1 adds a specialized decision-scoring layer rather than another general-purpose generator.
The planned rebases on MAI, OpenAI and other models matter because they separate the decision product from any one foundation model. That makes the base model an input to Microsoft’s decision workflow, not necessarily the product boundary.
First-order effects
Microsoft gains a fast decision-scoring offering built on Qwen3.5-9B, giving customers a model aimed at structured choices rather than text generation alone.
MAI and OpenAI models become planned alternative foundations for later Decision-1 versions, while Microsoft retains the decision-scoring product layer.
Second-order effects
Enterprise buyers can compare decision-task latency and calibrated scores alongside general-model quality, increasing pressure on model providers to support specialized workflow layers.
Microsoft’s existing business-focused model portfolio can be assembled around different base models, reducing the need to tie every workload to a single model family.
Third-order effects
If model-agnostic rebasing becomes standard, differentiation shifts toward evaluation, routing and workflow-specific tuning rather than exclusive ownership of the underlying foundation model.
AI procurement moves toward choosing models by task and measurable operating characteristics, giving platforms that can swap base models more leverage over deployment decisions.
The trend: Decision models are emerging as a distinct application layer in which the foundation model is interchangeable and speed, scoring quality and workflow fit become the buying criteria.
Microsoft-Decision-1 achieved the highest accuracy in our 36-benchmark comparison, spanning nearly 150,000 questions across benchmarks kept blind from training.
This is our first release in this new class of decision models, and we will continue to release updates including versions built on @MicrosoftAI and OpenAI models in the future. Get started with Microsoft-Decision-1 today in Microsoft Foundry and soon through OpenRouter. https://…
Microsoft-Decision-1 is live on OpenRouter. Microsoft's numbers: highest accuracy across 36 blind benchmarks (~150K questions), 4.5x faster than the runner-up, 35x faster than GPT-6 Sol, and decisions flip on just 1.3% of perturbed inputs. Post-trained from Qwen3.5-9B. $0.042/M i…
This is built on Qwen3.5-9B base model, comes with one capacity, and works only for text. Trained on somewhere between “a billion and a trillion” tokens (how's that for preciseness!).
Introducing Microsoft-Decision-1 Our model for fast decision-making that delivers top performance in latency and quality on structured decision tasks to outperform both LLMs and other decision models. https://commandline.microsoft.com/ ...
Introducing Microsoft-Decision-1, our new model for fast decision-making. It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality. We're already testing it across Microsoft for everything from incident res…
Classification is a key use case for decision models. Check out how accurately and quickly Microsoft-Decision-1 can categorize a variety of queries compared to GPT-6 Sol.
bold to send a message that you don't even need to benchmark against competitors when the stickiness of existing customers is so high that it wouldn't even be worth initiating procurement for a new vendor. well played!
NEW: Microsoft also releases its System One model, Microsoft-Decision-1. What a crazy effect Jev has had in the space. Also, a very cool application of decision-making models to power LLM judges and improve scientific discovery pipelines (e.g., screening candidate hypotheses). I'…
Microsoft launched Decision-1, a model built for decisions, not long conversations. The numbers are interesting: → 35x faster than GPT-6 Sol → 4.5x faster than the runner-up → Highest accuracy across 36 benchmarks and ~150K questions → 14x faster and 200x cheaper for Xbox feedbac…
Microsoft just dropped its own System One model: Microsoft-Decision-1. Jev is having a crazy impact on the AI space. The interesting part? Decision models could power LLM judges, evaluate outputs, and even help scientists screen candidate hypotheses. But there's still a big quest…
The myriad of decision models we see popping up fast and all over the place is a testament to the plentifulness and ease of getting synthetic data and fine-tuning awesome on top of awesome open-source models smaller than ~40B.
microsoft just walked into the decision model game typesafe's jev started it in september, cloudflare's clef followed, and now satya is launching microsoft decision 1 for fast structured decisions, claiming it beats both llms and the other decision models on latency and quality c…
We're going to see an explosion of decision models in the next six months. Decisions are inherently company specific and really need to be calibrated for specific preferences. We very low key launched training decision models using Applied Compute yesterday. Can't wait to see wha…
Meet Microsoft-Decision-1 by @Microsoft A small model built for one job: fast decisions Routing, classification, agent controls, AI judging. It returns a calibrated probability for every option. → Post-trained from Qwen3.5-9B → 35x faster than GPT-6 Sol → 4.5x faster than the run…
Official JevBench results: Microsoft-Decision-1 scores #6 in its peer group of API provided decision models. Compared with Jev it performs strong on many typical use cases, most notably on Recruiting, Risk assessment, Gaming, Advertising, Science and Moderation. It's also the str…
When sth breaks at 3am you dont need an essay... you need a fast decision Microsoft decision 1 is built for exactly that → beats LLMs and other decision models on speed and quality for structured decisions They're already testing it across microsoft Smart move tbh
Microsoft just shipped an AI that makes decisions faster than you can blink. A blink takes about 100 milliseconds. This one takes 85. And Satya didn't pick this day by accident. A few hours before this post went up, a startup called TypeSafe closed an $870 million Series A at a $…