Microsoft Launches Decision-1 AI Model Built on Alibaba's Qwen, Promises 35x Speed Boost

Microsoft Enters Crowded AI Decision Model Market

Microsoft has launched Microsoft-Decision-11

1

2

2

, joining over 100 companies competing in the emerging AI decision model category. Achint Srivastava, VP of software engineering in Microsoft's Office of the CTO, announced the model on October 9, describing decision models as "quickly emerging as an important new category in AI."3

3

Unlike large language models designed to generate text or reason through complex problems, Microsoft-Decision-1 delivers structured outputs that software can immediately act on, making it purpose-built for routing, classification, and verification tasks2

2

.

Built on Alibaba's Open Weight Model

Microsoft-Decision-1 is built on Alibaba Cloud's Qwen3.5-9B, an open weight model from the Chinese tech giant's cloud computing arm1

1

3

3

. Microsoft applied specialized post-training to improve decision scoring capabilities4

4

. However, Microsoft stated it will soon rebase Decision-1 on its own MAI models and OpenAI models, though the company did not disclose reasons for this planned transition1

1

3

3

. The model features a 32,000-token context window and accepts text-only inputs3

3

.

Performance Claims Show Significant Speed Advantage

Microsoft claims Microsoft-Decision-1 achieved the highest accuracy across 36 benchmark tests covering nearly 150,000 questions kept blind from training2

2

3

3

. The company reports median-latency performance approximately 35 times faster than OpenAI's GPT-6 Sol and 2.5 times faster than H2O-Lightning-4B3

3

4

4

. Microsoft-Decision-1 also demonstrated 2.8 times faster performance than Jev, the model from TypeSafe AI that launched three weeks earlier and became the fastest-adopted model on Vercel this year1

1

. When Microsoft tested consistency by rewording or reordering requests, the model changed its decision only 1.3% of the time on average3

3

.

Cost-Effective Alternative for High-Volume Decision-Making Tasks

Microsoft prices Microsoft-Decision-1 at $0.042 per million input tokens, with output tokens provided free of charge2

2

3

3

. According to Microsoft's calculations, a request using 86 input tokens and two output tokens costs approximately $0.0000036, meaning $1 would cover more than 275,000 similar decisions2

2

. Microsoft's Xbox Research team tested the model on more than 10,000 pieces of player feedback and reviews, finding it competitive with GPT-6 Sol on quality while operating more than 14 times faster and approximately 200 times cheaper3

3

4

4

. Srivastava emphasized that "now that agentic AI is a reality, we've seen that cost plays a major role in how people decide to use AI."1

1

How Decision Models Separate Decision-Making from Generative AI

Instead of generating free-form text responses, Microsoft-Decision-1 takes a fixed set of answer options and returns calibrated probabilities for each one in a single pass2

2

. The model handles yes-or-no judgments, multiple choice selections, ratings, and rubric-based assessments of AI responses and agent actions2

2

3

3

. Charlie Dai, vice president and principal analyst at Forrester, noted that "routing, classification, and approval checks often represent a large share of agent interactions, so a specialized AI decision model can significantly reduce cost and latency while reserving general-purpose LLMs for reasoning and content generation."2

2

For agentic AI workflows, Microsoft-Decision-1 can determine whether tasks should proceed, require repetition, escalate to higher priority, or need human intervention4

4

.

Applications Span Content Classification to Workflow Control

Microsoft lists use cases including content classification, request routing, incident prioritization, quality control, outcome validation, and safety checks2

2

4

4

. The model can evaluate multiple decisions in parallel within a single request, making it suitable for high-volume structured business decisions2

2

. Microsoft's Copilot team found Decision-1 competitive with GPT5.6 Luna while being 100 times faster3

3

. The model is available through Microsoft Foundry and OpenRouter, the model router that Stripe acquired in August3

3

.

Governance Risks Require Careful Implementation

Microsoft-Decision-1 returns calibrated probabilities but does not provide reasoning or explanations for its decisions2

2

. Amit Kumar Jena, development manager for AI at Kanerika, cautioned that "a precise-looking probability can give an uncertain decision more authority than it deserves."2

2

Microsoft's documentation warns that results can change depending on how questions are worded2

2

. Dai recommended calibration testing, controls over question wording and score thresholds, audit logging, and human review for high-impact decisions, particularly when customer outcomes or regulatory obligations are involved2

2

. The model card explicitly states it should not be the sole basis for consequential decisions about people, such as credit or employment determinations3

3

. Microsoft tested Decision-1 across 5,250 requests covering 11 safety benchmarks, including harmful content identification, jailbreak attempts, and prompt injection scenarios3

3

4

4

.