Clef vs Granite Switch 4.1

Clef

Apache-2.0 weights for all three models, ungated on Hugging Face. Clef-omni (9 October 2026) adds audio and video input. Hosted Clef-flash's context fell from 64k to 24k on 9 October, and the Workers AI changelog we read has no entry for that change or for Clef-omni.

Granite Switch 4.1

One self-hosted Apache 2.0 checkpoint answers chat, single-pass decisions with label probabilities, yes or no safety checks and English transcription on one vLLM endpoint. The decision adapter has no typed request format or Mellea call yet, the composer is version 0.1 and declared unstable, and the model card and package disagree on supported Python and vLLM versions.

Clef

1. Send model in the body as clef, clef-flash or clef-omni, matching the model in the URL. The input schema requires it
2. Ask every independent question about one state in one call. Up to 64 are allowed
3. Try clef-flash first for text-only triage under 24k tokens. It costs $0.038 per million input tokens against $0.24 for Clef
4. Send Clef-omni audio and video as base64 data URIs in audio and videos. Remote URLs are not accepted
5. Media and questions count towards the context window. If they exceed it, the request fails. Otherwise text state is truncated to fit
6. Read the internal code on a 429. 3036 means the day's 10,000 free neurons are spent, 3040 means capacity, so retry later
7. Treat an answer about user-supplied text, images, audio or video as a judgement that hostile input can steer. Cloudflare publishes no guidance on it

Granite Switch 4.1

1. Pass chat_template_kwargs: {"adapter_name": "decision"} or "guardian-core". Without it the request is ordinary chat on the base model
2. For a decision, send the state and the ### DECISION question as two user turns, request logprob_token_ids for the option labels, and set max_completion_tokens to 1
3. Use single-token option labels. Split more than 128 labels into chunks with the identical prompt and apply one softmax over all returned logprobs
4. Parse guardian-core answers as JSON, {"score": "yes"} or {"score": "no"}, where yes means the criterion is met
5. Add authentication before exposing the vLLM server. Neither vLLM nor the checkpoint requires a key by default