Cloudflare and AWS Jump Into Open Decision Models With Clef and Strands Decider 2B
On October 1, Cloudflare launched Apache 2.0 Clef 27B and Clef-flash 9B with a Jev-compatible API and RL fine-tuning pipeline, while AWS open-sourced Strands Decider 2B.
Two weeks after Jev’s debut, two major cloud infrastructure providers released open-weight decision models on the same day, October 1: Cloudflare unveiled Clef (27B) and Clef-flash (9B) on Workers AI, while AWS released the compact open-weight Strands Decider 2B.
Drop-in Jev compatibility plus RL fine-tuning
Built on frozen Qwen backbones and published under the Apache 2.0 license on Hugging Face, Clef and Clef-flash were the first models trained in-house by Cloudflare’s Workers AI team. Critically, Cloudflare designed both endpoints to match TypeSafe’s System One API format—supporting up to 64 questions, 64K context, and four images per request—so developers could migrate simply by changing the endpoint URL and model ID.
Cloudflare Clef & Clef-flash
27B and 9B Apache 2.0 models with vendor-reported median latencies of 209.3 ms (Clef) and 38.8 ms (Clef-flash), paired with an AI Gateway → Workers AI → Trainer RL fine-tuning loop.
AWS Strands Decider 2B
A compact 2-billion-parameter open model targeting ultra-cheap agent routing and tool-selection hops at the edge of the AWS Strands SDK ecosystem.
In vendor benchmarks, Cloudflare reported Clef leading on seven of ten tested decision benchmarks (while Jev held leads on When2Call and BRIGHT). Eight days later, Cloudflare cut Clef-flash to $0.038 per million input tokens and introduced multimodal Clef-omni.