Cloudflare announced the Clef family on Thursday, two weeks after Jev's release. The lineup consists of Clef and Clef-flash, a slightly smaller and faster version. Like TypeSafe's Jev, both answer three types of bounded, structured questions: yes/no, multiple choice and rankings.
Architecturally, Clef uses specially post-trained, frozen versions of Qwen3.8-27B and Qwen3.5-9B for Clef and Clef-flash respectively, with the Qwen backbone performing a prefill-only pass during inference. Cloudflare says Clef is faster than Jev and scores choices in parallel after that pass. Jev's underlying architecture remains undisclosed.
On benchmarks, Cloudflare ran Clef against Jev and other open decision models using the Jev Decision Index on Hugging Face. The company's own ranking suggests Clef is slightly slower than other open models but more accurate, while Clef-flash is as accurate as most others and far faster. Cloudflare also tested Clef against TypeSafe's own benchmarks, claiming it beat Jev in three of four areas, losing only on agent trace observability. These results are self-reported and have not been reproduced for the official Decision Index.
Clef's capabilities extend beyond text to images and video, with a 64k context window. Jev also handles up to 64k tokens per request, though its state plus longest individual question is limited to 32k.
Clef is hosted on Workers AI, where Cloudflare said it can exploit edge GPUs: "We're able to take advantage of our GPUs at the edge, leading to low network latency and faster decisions." Pricing is $0.24 per million tokens, nearly six times Jev's $0.042 per million. The models are downloadable from Hugging Face under Apache-2.0 terms, though Cloudflare AI Platform group product manager Michelle Chen confirmed training datasets are not public. Clef-flash runs on any GPU with at least 41 GB of VRAM, while Clef requires 85 GB, "assuming single concurrency and a 64k context window," Chen said. The API is fully Jev compatible, allowing Clef to serve as a drop-in replacement.