Microsoft skipped OpenAI’s decision model and built its own on Alibaba’s Qwen
Microsoft on Friday launched Microsoft-Decision-1 in Foundry, three days after OpenAI opened its Decisions API to every developer in public The post Microsoft skipped OpenAI’s decision model and built its own on Alibaba’s Qwen appeared first on The New Stack.
Microsoft launched its own decision-making AI model, Microsoft-Decision-1, built on Alibaba's Qwen3.5-9B instead of OpenAI's framework, positioning itself against competitors like TypeSafe and OpenAI.
What happened
Microsoft introduced Microsoft-Decision-1, a decision-making AI model trained on Alibaba's Qwen3.5-9B, as part of its Foundry platform. The model competes with OpenAI's Decisions API and TypeSafe's Jev, offering faster processing and lower costs. Microsoft also highlighted its internal use cases and plans to rebase the model on its own infrastructure.
Why it matters
The model reflects Microsoft's strategy to reduce reliance on third-party AI providers and optimize costs for enterprise applications. Its integration into critical workflows like Copilot and Xbox underscores its role in shaping AI-driven decision-making systems.
- Microsoft prioritizes in-house AI development to reduce dependency on external providers.
- Cost and speed are critical factors in enterprise AI adoption.
- Decision models are becoming a competitive battleground with diverse technical approaches.
- Interoperability and open standards remain unresolved challenges in AI ecosystems.
- Adversarial testing and calibration are critical for trust in AI decision systems.
The bigger picture
Microsoft's decision to build its own model may influence AI infrastructure choices for enterprises, while its focus on cost efficiency and internal integration highlights trends in AI adoption. The lack of image support and open weights could limit its appeal compared to competitors.
Microsoft's launch of Decision-1 marks a strategic shift in AI development, leveraging third-party models while planning future in-house advancements. Its focus on speed, cost, and internal integration highlights evolving priorities in enterprise AI.
Technical details
- Model base: Alibaba's Qwen3.5-9B
- Token capacity: 32,768 tokens of text, JSON output
- Pricing: $0.042 per million input tokens
- Calibration: Probabilities are calibrated for reliability, though adversarial testing remains unverified
- Architecture: Designed for structured decision tasks with low latency
FAQ
Why did Microsoft use Alibaba's Qwen instead of OpenAI's models?
Microsoft chose Qwen3.5-9B as a foundation for Decision-1, though it plans to rebase the model on its own MAI infrastructure in the future.
What are the key technical limitations of Microsoft-Decision-1?
The model lacks image support, does not provide open weights, and has unverified performance under adversarial conditions.
How does Decision-1 compare to competitors like Jev and Clef?
Decision-1 offers faster processing and lower costs but lacks image handling and open-source availability compared to some rivals.
This summary and analysis were generated with AI from the original article and may contain errors. Read the original for the full story.
Read the full story
Originally published by The New Stack - AI on Oct 10, 2026.
Comments
Sign in to join the discussion.