Microsoft releases Qwen3.5-9B decision model Decision-1

Microsoft released Decision-1, a Qwen3.5-9B model that scores fixed options, claiming 83.5% accuracy at 85 ms.

Microsoft released a decision-scoring model, named Microsoft-Decision-1 by MarkTechPost and Decision-1 by The Decoder, post-trained from Alibaba’s Qwen3.5-9B. It returns a calibrated probability for each fixed answer option rather than generated text, aimed at routing, classification, verification, evaluation, and agent control. Microsoft says it led 36 benchmarks covering nearly 150,000 questions, with 83.5% accuracy at 85 ms latency, and ran 2.5 times faster than H2O-Lightning-4B, ahead of Jev 1.13.0. Cloudflare’s Qwen-based Clef models were excluded from that comparison. The model is available in Microsoft Foundry and on OpenRouter at $0.042 per million input tokens, with output tokens free.

• Microsoft released a decision model called Microsoft-Decision-1 by MarkTechPost and Decision-1 by The Decoder, post-trained from Alibaba’s Qwen3.5-9B.
• It returns a calibrated probability for each fixed answer option rather than generated text, for routing, classification, verification, evaluation, and agent control.
• Microsoft says it led 36 benchmarks covering nearly 150,000 questions, scoring 83.5% accuracy at 85 ms latency.
• Microsoft says it runs 2.5 times faster than H2O-Lightning-4B and is ahead of Jev 1.13.0; Cloudflare’s Qwen-based Clef models were excluded from the comparison.
• It is available in Microsoft Foundry and on OpenRouter at $0.042 per million input tokens, with output tokens free.
• Both reports are dated 2026-10-10.