The decision-making model market is growing rapidly, with Cloudflare recently introducing two new tools: Clef and Clef-flash. These follow the launch of Jev by TypeSafe, another company in the field. Unlike traditional large language models (LLMs), which are often unpredictable and require significant computing power, decision-making models are designed for specific tasks. They provide structured answers—like yes/no, multiple choice, or rankings—along with probability scores. These models are particularly useful in automated systems where accuracy and reliability are essential. Clef is built on frozen and post-trained versions of the Qwen3.8-27B and Qwen3.5-9B models and uses a non-autoregressive method, which allows for faster processing. Clef has an advantage over Jev because it can analyze images and videos, while Jev is limited to text. Clef also has a much larger context window, capable of handling 64,000 tokens of information. However, using Clef on Cloudflare's hosted infrastructure is more expensive than using Jev—0.24 dollars per million tokens versus 0.042 dollars for Jev. Companies can run Clef locally under the Apache 2.0 license, but this requires powerful hardware. Clef-flash, a lighter version, can run on GPUs with at least 41 GB of VRAM, while Clef needs 85 GB of VRAM. For business leaders, the appeal of Clef lies in its ability to handle repetitive tasks efficiently without compromising accuracy. For instance, Cloudflare's Threat Intelligence team uses Clef to classify web domains. This process is significantly faster than using traditional LLMs, cutting latency in half and helping detect phishing sites more quickly. Clef can also analyze incoming client requests, determine their urgency, and route them to the appropriate department—technical, billing, or commercial—with consistent reliability. Alongside Clef, Cloudflare also launched K2, a serverless event streaming service currently in public beta. K2 solves a common challenge in distributed systems: separating data producers from consumers. While many companies use complex setups like Apache Kafka clusters, K2 uses Cloudflare's R2 object storage, offering high scalability and reliable long-term data retention. However, this approach introduces a slight delay in processing, about one second at the 99th percentile, due to batch processing. K2 is currently free in beta, but the company plans to introduce a competitive pricing model when it officially launches, with costs of 0.04 dollars per GB for both produced and consumed data, and 0.02 dollars per GB per month for data retention. By combining K2 with Clef, Cloudflare provides IT directors with a practical, high-performance solution that avoids the unpredictable nature of generative AI. This approach allows for fast, deterministic decision-making directly on Cloudflare's global network infrastructure, offering a balanced alternative to the current enthusiasm for more general AI tools.