Microsoft Adopts Alibaba’s Qwen Model for New Decision AI
Microsoft has launched Microsoft-Decision-1, a specialized decision model built initially on Alibaba Cloud’s Qwen3.5-9B architecture, to challenge TypeSafe AI’s Jev model in the rapidly expanding enterprise automation market.
The Shift Toward Specialized Decision Models
Large language models excel at generative tasks, but they introduce unpredictability and high compute costs when deployed for deterministic software routing. Decision models operate differently. TypeSafe AI released Jev three weeks prior, introducing an LLM variant tuned to respond to specific questions with probability-rated constraints rather than open-ended text generation. These systems minimize conversational drift and eliminate generative hallucinations, though they remain vulnerable to classification errors.
OpenAI introduced its Decisions API into public beta, while Cloudflare released Clef. Other new entrants include Strands with its Decider 2B, Liquid AI with d1, Perplexity’s Decisions API, Snowflake’s decision model, Surogate Rune, and H2O.ai with H2O-Lightning-4B, which also utilizes a Qwen base model. More than 100 decision models now compete for enterprise workloads.
Performance Metrics and Architectural Roadmap
Achint Srivastava, VP of software engineering in the Office of the CTO at Microsoft, detailed the engineering rationale in a blog post. Srivastava stated that decision models are purpose-built to deliver structured outputs that software can immediately act on, unlocking useful automation tasks at very low cost with high performance.
Microsoft-Decision-1 is currently accessible via Microsoft Foundry and will soon appear on OpenRouter. The underlying architecture relies on Alibaba Cloud’s Qwen3.5-9B, though Microsoft plans to rebase the model onto its own proprietary infrastructure and OpenAI models in subsequent releases.
Internal Microsoft latency tests place Decision-1 at 2.5 times faster than H2O-Lightning-4B and 2.8 times faster than Jev. The model achieved an 83.5 percent accuracy rate across 36 benchmarks and secured a 92.2 percent confidence score, ranking second behind Quyet-1.0-Large.
Pricing Structures for Agentic Workflows
Cost efficiency drives enterprise adoption in agentic AI deployments. Microsoft-Decision-1 charges $0.042 per million input tokens, while output tokens are free. Srivastava noted that cost plays a major role in how organizations decide to deploy AI at scale.
Microsoft claims its new model undercuts OpenAI’s GPT-6 Sol by more than 20 times in standard text classification tasks. As competition across the 100-plus available decision models intensifies, architecture and inference costs will dictate which platforms capture enterprise market share.
