The Pulse
Cloudflare’s Clef Decision Model Arrives on Ollama Cloud
Ollama announced that Cloudflare’s 27-billion-parameter Clef decision model is available through its cloud service and for local use. The announcement’s 10 benchmark rows show Clef ahead of Jev on eight benchmarks and behind on When2Call an
AI.info Team · Written with AI · fact-checked ·
Ollama made Cloudflare’s 27-billion-parameter Clef decision model available on Ollama Cloud on October 19, giving users a hosted option for a model that returns structured decisions instead of open-ended text. Clef can assess text, JSON and images, then assign probabilities to answers defined by the caller. Ollama also lists it for local use.
Clef returns scores, not a drafted reply
A conventional language model generates text token by token. Clef instead takes an input state and typed questions, then scores possible answers together in a single pass. An application can ask whether a support ticket is urgent, which team should handle it, or whether a proposed action meets a defined policy.
The output is designed for software to act on directly: a selected option or yes-or-no result, alongside probabilities for the answers. That makes Clef suited to bounded classification and routing tasks, rather than requests that need a written explanation or broad reasoning. Ollama says the model can also evaluate images alongside text, including screenshots, receipts, forms and photos.
Cloud access comes with version and plan requirements
Ollama’s model listing says cloud use requires Ollama version 0.40.6 or later; the locally run model requires version 0.35.1 or later. Cloud use requires a paid plan or purchased usage credits, and does not require downloading the model weights. The listing also describes Clef as fine-tuned from Qwen3.8-27B and released under the Apache 2.0 license.
Ollama’s announcement says Clef is compatible with the Jev and System One APIs. The model listing documents use of Ollama’s /v1/systemone endpoint and shows a cloud invocation using the model name clef:cloud.
Ollama’s benchmark table shows a mixed comparison
Ollama’s announcement reproduces results from Cloudflare’s evaluations against Jev and other models. Across the 10 benchmark rows it displays, Clef scores higher than Jev on eight, including BFCL case exact, ToolRet nDCG@10, API-Bank accuracy, the home-appliances case-exact test, BANKING77 macro-F1, CLINC150+OOS macro-F1, Amazon ESCI macro-F1 and PhishNChips accuracy. Clef scores lower on When2Call accuracy, 72.37 to Jev’s 80.97, and BRIGHT nDCG@10, 45.91 to Jev’s 47.52.
The figures are benchmark results presented by the companies, not an independent evaluation of Clef in Ollama Cloud. They also measure specific tasks; they do not establish that Clef is more accurate for every classification workflow or that its hosted response times match results from another service. Ollama’s announcement provides no Ollama Cloud latency figure.
A narrow tool for defined decisions
Clef’s value depends on whether a developer can express a task as a set of typed questions and permitted outcomes. When that fits, its probabilities can feed a routing rule, policy check or tool-call review without parsing a generated paragraph. When the task requires open-ended explanation, Clef is not a substitute for a general-purpose language model.
The release puts Cloudflare’s model within Ollama’s cloud offering as well as its local model library. For hosted users, the practical entry points are a compatible Ollama version and a paid plan or usage credits; for teams measuring accuracy, the published benchmark table includes clear exceptions on When2Call and BRIGHT.