Cloudflare launches open-source Clef models for structured AI decisions
Clef and Clef-flash are available on Workers AI, while customers seeking tailored versions can use Cloudflare’s fine-tuning team.
Cloudflare has released Clef and Clef-flash, open-source decision models designed to return structured, probability-based outputs for classification and agent workflows. The company says the models can help software route requests or trigger actions based on defined choices.
Both models are hosted on Cloudflare’s Workers AI platform, and their weights are available on Hugging Face under an Apache 2.0 licence. Cloudflare says Clef is intended for more demanding decisions, while Clef-flash is designed for latency-sensitive uses.
The launch also includes a reinforcement learning service to fine-tune Clef for customer workloads. Cloudflare is initially offering this through its hands-on engineering team; it plans to develop a self-service platform, but gave no timing.
Cloudflare reported that Clef classified a website in 2.2 seconds in a company test, compared with 4.7 seconds for gpt-oss-120b in the same workflow. The company also said Clef returned more classifications. These are Cloudflare’s own comparisons, with no independent validation provided in the announcement.
Cloudflare says the models are based on Qwen and post-trained for decision tasks. Its approach scores valid choices against a defined schema, allowing the model to return typed outputs and probabilities rather than generate an open-ended response.
