Physix Frontier · News Briefing Card (IT Home · Oct 10, 2026)

Microsoft Launches Decision-1 Model, Post-Trained on Qwen3.5-9B

KEY FACTS

Microsoft launched the Microsoft-Decision-1 decision model on October 9, designed specifically for structured decision-making tasks.
The model is now available on Microsoft Foundry and has also been integrated into the OpenRouter platform.
In 36 benchmark tests, the model achieved the highest accuracy and ranked first in response speed.
Its response speed is 4.5 times faster than Quyet-1.0-Large and 35 times faster than GPT-6 Sol.
Microsoft post-trained Qwen3.5-9B and will later migrate to base models such as MAI and OpenAI.

KEY DATA

$0.042 per million tokensInput price

FreeOutput price

36Number of benchmark tests

5,250Number of safety test requests

PHYSIX OBSERVATION

Microsoft's use of the Qwen base for post-training shows that open-source models are becoming a shortcut for big tech to quickly customize specialized capabilities. The decision model focuses on low latency and low cost, and free output is more likely intended to seize the entry point of the Agent decision layer. For the industry, beyond general large models, vertical decision small models may become a new competitive point.

Source: IT Home report