Cloudflare Launches Clef: Open-Source Decision Models & RL Platform
Original: Clef: Open-source decision models, and new RL fine-tuning platform
Why This Matters
Bringing RL fine-tuning to edge-native AI tooling lowers the bar for specialized model adaptation outside major cloud providers.
Cloudflare has introduced Clef, an open-source suite of decision models and a reinforcement learning fine-tuning platform. The release, announced during Birthday Week, is aimed at developers building AI-powered decision workflows on Workers AI and the broader Cloudflare developer platform.
Cloudflare has unveiled Clef, a new open-source project combining decision-focused AI models with a reinforcement learning (RL) fine-tuning platform. The announcement came as part of Cloudflare's Birthday Week batch of product releases.
Decision models differ from typical generative models in that they are optimized for classifying, routing, or acting on inputs rather than producing long-form text. Clef is designed to make these models accessible to developers via Workers AI, Cloudflare's inference-at-the-edge offering.
The RL fine-tuning platform component is notable: it lets developers adapt base decision models to specific tasks using reward signals, a method that has powered improvements in models like OpenAI's o-series. By open-sourcing Clef, Cloudflare positions the tooling for community contributions and transparent evaluation. The company has not disclosed specific benchmark numbers or model sizes in the blog post summary available, but the release aligns with Cloudflare's broader push to own more of the AI developer stack—from inference to training feedback loops—within its edge network.