Microsoft on Sunday introduced Microsoft-Decision-1, a decision-scoring model trained on Qwen3.5-9B and designed for fast classification, routing, and workflow control tasks, the company said.
The model targets growing demand for lower-cost, lower-latency alternatives to large language models for high-volume structured decisions. Decision models produce calibrated confidence scores for predefined options rather than open-ended text.
Microsoft-Decision-1 is the result of post-training Alibaba's Qwen3.5-9B for single-pass decision scoring, according to the company. Microsoft said it will rebase future iterations on its own MAI models and OpenAI models. The model is now available in public preview through Microsoft Foundry and OpenRouter, with input tokens priced at $0.042 per million and output tokens free.
In a 36-benchmark evaluation spanning nearly 150,000 questions kept blind from training, Microsoft-Decision-1 achieved the highest accuracy, the company reported. It was 2.5 times faster than H2O-Lightning-4B v1.1, the runner-up, and 35 times faster than GPT-6 Sol on a P50 latency basis. Microsoft also said the model's robustness to input perturbations kept the decision flip rate at 1.3 percent on average, with zero flips when option descriptions were paraphrased or options reversed or shuffled.
Xbox Research used Microsoft-Decision-1 to classify more than 10,000 pieces of open-ended feedback, finding quality competitive with GPT-6 Sol while running over 14 times faster and 200 times less expensive, Microsoft said. The Copilot team reported the model was competitive with GPT5.6 Luna and 100 times faster for response quality assessments. In scientific discovery, an adaptive replanning scenario produced scores 46 times more consistent than an LLM-based scorer at three times the speed, resulting in nearly four times faster adaptive replanning overall.
The company did not specify when the model would move out of public preview.