Cloudflare: Clef Flash

Cloudflare: Clef Flash is a decision model from Cloudflare, the company behind Workers AI and a broad edge network. It belongs to Cloudflare's open source Clef family, and this flash variant is the fast 9B member of that family. It is a fine tune of Qwen3.5-9B and is served on Workers AI, so it sits within Cloudflare's own inference stack rather than being a generic hosted checkpoint.

The model takes a state as input, which can be plain text or structured JSON, and it also accepts images. Its output is a decision rather than free form prose, which makes it a different kind of tool from a general chat model. The context window is 65,536 tokens, and the maximum output is 14,745 tokens. Because the output modality is decisions, you should expect the model to return a choice or classification style result that your application then acts on, not a long written answer.

That shape points to a fairly specific set of uses. Anywhere your code needs to pick between options based on a mix of text and structured data, Clef Flash is a natural fit: routing or triage logic, classifying an incoming request, deciding whether an item matches a rule set, or choosing the next step in a workflow. The image input extends this to cases where the state includes a screenshot, a document scan or a photo alongside text or JSON. The JSON input support matters too, since you can pass a structured state directly instead of flattening everything into prose.

You call it through api14's OpenAI compatible API using the model id "cloudflare/clef-flash". Because the endpoint follows the OpenAI schema, you can point existing OpenAI client libraries at api14 and change the model id rather than rewriting your integration. Note that this model lists no supported request parameters, so you should not rely on tuning knobs such as temperature or top_p being honoured; send your state in the standard message format and read the decision from the response.

The main thing to weigh is the output modality. If your task needs generated text, explanation or conversation, a decision model is the wrong shape, and a general text model will fit better. Clef Flash is also a 9B model, so it is built for speed rather than for the heaviest reasoning workloads. Use it when you want a fast, cheap to run decision step over text, JSON or images within a 65,536 token context, and look elsewhere when the job is producing prose.