Signal

Cloudflare releases open Clef decision models and RL fine-tuning

Clef and Clef-flash are hosted on Workers AI with Apache 2.0 weights; Cloudflare also introduced a reinforcement-learning fine-tuning service.

1 min read
Cloudflare's Clef and Clef-flash launch post and reinforcement-learning fine-tuning announcement
Cloudflare announces its open-source Clef decision models and RL fine-tuning service · Credit: Cloudflare View source

Cloudflare announced Clef and Clef-flash on October 1 as its first decision models trained by the Workers AI team. The models take an input state and typed questions and return probabilities over allowed answers rather than free-form text.

Hosted access, open weights and tuning

Both models are available on Workers AI, and Cloudflare says it is releasing their weights under Apache 2.0. The company also opened a reinforcement-learning fine-tuning service for customers adapting Clef to specific workloads.

Cloudflare’s speed and benchmark comparisons are vendor-run results. The announcement reports 43 latency runs and 10 decision benchmarks, but independent reproduction was not located in this review.

Sources