Microsoft leans on open weight model from Chinese AI lab to challenge Jev
Microsoft has launched Microsoft-Decision-1, a low-cost model designed to make structured classifications and decisions, joining a fast-growing field spurred by attention around TypeSafe AI’s Jev. Such models can give software direct, constrained outputs for business tasks, though they can still make errors.
The first version is built on Alibaba Cloud’s Qwen3.5-9B; Microsoft says a future release will use its own and OpenAI models. The company claims 83.5 per cent accuracy across 36 benchmarks, a 92.2 per cent confidence score, and latency 2.5 times faster than H2O-Lightning-4B and 2.8 times faster than Jev in its test. Input costs $0.042 per million tokens, with output free, and more than 100 decision models are competing for attention.
- Microsoft has introduced a decision model built on Alibaba’s Qwen3.5-9B.
- It claims 83.5 per cent accuracy across 36 benchmarks.
- Microsoft says a later version will use its own and OpenAI models.
New here? Start with this
Structured decision-making models are artificial intelligence systems designed to analyse information and make specific, constrained choices for business tasks. Rather than generating open-ended text, they provide direct outputs—such as classifications or yes/no decisions—that software can act on reliably, reducing the risk of unpredictable AI behaviour. However, these systems can still make errors.
Microsoft's new model, Decision-1, is part of a rapidly expanding market for these specialised tools. The company built its initial version using technology from Alibaba Cloud, a Chinese technology firm, but plans to offer versions using its own and OpenAI's technology in future. Other companies, including TypeSafe AI with its model Jev, are also competing in this space.
The growth of this field reflects business demand for AI that can handle routine decision-making tasks efficiently and cost-effectively. These models are cheaper to run than larger general-purpose AI systems, which makes them appealing for companies handling high volumes of structured decisions. The competition between providers suggests the market will remain active as organisations seek reliable, affordable solutions.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Microsoft's competitive approach offers genuine benefits through lower costs and faster inference whilst maintaining a healthy market with over 100 alternatives. Using proven open-weight models represents pragmatic engineering that drives innovation, and a genuinely competitive landscape naturally prevents any single supplier from exercising problematic dominance.
The case against
Supply chain resilience and geopolitical risk warrant serious consideration when critical business infrastructure depends on Chinese-origin AI systems, particularly as global tensions escalate and export controls tighten. The competitive drive to cut costs and improve speed may sacrifice the robustness and safety validation these systems need for high-stakes applications, where 83.5% accuracy still represents meaningful error rates.