Microsoft leans on open weight model from Chinese AI lab to challenge Jev

Microsoft has joined the Jev fan club, a group of companies that have all developed their own decision models. Jev, a large language model (LLM) from TypeSafe AI, is designed to provide responses to specific types of questions, with ratings based on probability. Its speed, affordability, and response constraints make it well-suited for business applications where open-ended text with uncertain accuracy could be problematic.

One benefit of Jev is that decision models do not "hallucinate," or generate inaccurate information, as standard LLMs might. However, decision models can still make errors, prompting researchers to investigate how these errors might be amplified.

Jev's announcement inspired other companies to declare they have decision models of their own, despite machine learning researchers having long known how to create classifier models for probability-based decisions. Microsoft is now entering the race with its own decision model called Microsoft-Decision-1. Built on the Qwen3.5-9B model developed by Alibaba Cloud, a Chinese tech giant, Microsoft-Decision-1 is offered through Microsoft Foundry and soon OpenRouter.

According to Microsoft, their model is 2.5 times faster than H2O-Lightning-4B and 2.8 times faster than Jev in latency tests. Microsoft-Decision-1 also leads in accuracy (83.5 percent) on 36 benchmarks and ranks second in confidence score (92.2 percent) behind Quyet-1.0-Large. Input tokens cost $0.042 per million tokens, with output tokens being free.

The decision to invest in decision models is driven by the cost factor, as agentic AI becomes a reality. Srivastava emphasized the importance of choosing the right model for the right task, as cost plays a significant role in determining AI usage.