Microsoft skipped OpenAI’s decision model and built its own on Alibaba’s Qwen
The New Stack Amanda Caswell ● Covered by 7 sources
Microsoft launched its own decision model in Foundry, built on Alibaba’s Qwen. It’s a quiet shot at OpenAI and a way to keep more agent traffic inside Microsoft.
Based on reporting by The New Stack, Amanda Caswell — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Microsoft used Friday’s launch to send a very specific message: it doesn’t have to wait around for OpenAI. Microsoft-Decision-1 arrived in Foundry just three days after OpenAI opened its Decisions API to every developer in public beta, but Microsoft built its model on Alibaba’s Qwen3.5-9B instead of using anything from its partner. The company says it plans to rebase the model soon on its own MAI models and on OpenAI’s too. For now, though, the foundation is Alibaba’s.
The pricing is already familiar. Microsoft Decision-1 costs $0.042 per million input tokens with free output, the same rate TypeSafe charges for Jev. That puts Microsoft right in the middle of a suddenly crowded field where the economics are getting tight and the real prize is traffic, not token margin. TypeSafe says nearly 30% of the Fortune 500 has tried Jev, and CEO Diogo Almeida said on X that “29.4% of the Fortune 500 showed up” in the three weeks after launch. Microsoft sells Azure to those same companies, which makes that traction awkward to ignore.
Satya Nadella said Microsoft is already testing Decision-1 across the company, and he pointed to four internal uses. Xbox Research used it to sort more than 10,000 pieces of player feedback. The Copilot team used it to grade chat and agent responses. On-call engineers used it during live incidents to pull context. Microsoft Discovery used it to score experiments before an agent replans. Microsoft says the model is more than 14 times as fast as GPT-6 Sol for Xbox at a fraction of the cost, and 46 times more consistent in Discovery.
But the bigger play may be routing. Microsoft said on Wednesday that GitHub Copilot will soon decide when to run a task on-device and when to send it to cloud-scale models, though it hasn’t said what gets sent up to the cloud. Decision-making is one of the use cases Microsoft lists for Decision-1, and keeping those calls in-house would fit the company’s broader incentive across Copilot, GitHub, and Xbox. That incentive gets stronger as decision models get cheaper and more interchangeable.
There are still gaps. Microsoft’s Foundry listing says Decision-1 accepts up to 32,768 tokens of text and returns JSON, but it doesn’t support images. It also hasn’t announced open weights, and it hasn’t said whether the model is fully compatible with TypeSafe’s System One API, even though its sample code calls a /systemone endpoint. Microsoft says the probabilities are calibrated, but its own documentation tells customers to validate that on their own data. In a category moving this fast, that caution feels less like legalese and more like a warning label.
My take — AI-written commentary, not fact-checked reporting
Microsoft is doing what big platform companies always do when a new layer looks useful: it’s pulling the part that matters in-house and calling it strategy. Open standards sound nice until the routing logic sits between your users and your cloud bill. Then everybody discovers they’ve become very interested in “trust” and “compatibility.”
Read more about this at: The New Stack