Introducing Clef: our open-source decision models, and new RL fine-tuning platform

Cloudflare has released Clef and Clef-flash, two open-source decision models hosted on Workers AI that produce bounded, typed classifications and calibrated probabilities for automated workflow decisions such as support routing, domain categorization, and security triage. Positioned as alternatives to open-ended, non-deterministic large language models, the models are Jev API-compatible, support structured outputs, and are designed for fast, programmatic use in agentic systems; Clef also supports image inputs and has a 64,000-token context window. Cloudflare reports that Clef led the Jev Decision Index and that both variants outperformed several competing decision models on many accuracy and latency evaluations, with Clef-flash targeting especially latency-sensitive applications. Built on Qwen backbones, the models score predefined schema choices directly rather than generating responses token by token, using specialized attention routing, probability-calibration training, and reinforcement-learning-based optimization. The company is making the models available under an Apache 2.0 license through Hugging Face and Workers AI, while also launching a fine-tuning service that initially involves Cloudflare engineers and is intended to evolve into a self-service reinforcement-learning platform using AI Gateway, Containers, Trainer, and bring-your-own-model deployment tools.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.