Dongcha Beating AI News, after the release of Jev, AWS, Cloudflare and Databricks have recently followed up on the decision-making model.
AWS launches open source Strands Decider 2B. It is based on Qwen3.5-2B, removes the text generation module, replaces it with a decision-making module specifically responsible for selecting answers, and then uses LoRA to fine-tune it. In the official test, the median delay of a single decision on RTX 3090 was about 115ms, and the complete training took about 11 hours.
Cloudflare released Clef and Clef-flash in one fell swoop, based on Qwen3.8-27B and Qwen3.5-9B respectively, and both are open source. In Cloudflare's testing of Jev's official review set, the Clef series outperformed Jev in 3 out of 4 categories of tasks; Clef-flash's median latency was 38.8ms, while Jev's was 524.1ms.
Databricks has built this capability directly into the platform. The newly launched ai_decide can be called in SQL or REST API to classify, score and agent route data. The interface is also directly compatible with TypeSafe AI API. Officials say a decision can be returned in less than a second.
From developers themselves to forking it, to AWS, Cloudflare, and Databricks following up, decision-making models are rapidly changing from a new concept to a common component in Agent infrastructure.