Shattered, a tech news site, published an explainer on 11 October that adds engineering detail to Microsoft’s Decision-1 launch, which we covered on 12 October. Writer Tom Vance says the model is built on Qwen3.5-9B, Alibaba’s open-weight model with 9 billion parameters, and gives each choice on a preset menu of answers a probability rather than writing text. This is a third-party write-up, not Microsoft documentation.
According to Shattered, the model takes text only, accepts inputs of up to 32,000 tokens, and offers no explanations for its picks. Its worked example shows a priority score returned with 98.08% confidence, and applications are meant to act on that number, such as paging an engineer above a cutoff. Shattered also says the weights are not distributed, citing Data Studios, and that Microsoft has published no migration timeline for its planned move off Qwen.
A cutoff tuned against one base model may behave differently after a swap. Teams adopting Decision-1 now should save their calibration results so they can rerun them when the rebase arrives.
Reported by Shattered on 11 October 2026.
