KeyoAPI
Cloudflare Clef API — Multimodal Decisions on KeyoAPI
Cloudflare Clef is an open-source 27B multimodal decision model fine-tuned from Qwen3.8-27B.
Overview
Cloudflare Clef is an open-source 27B multimodal decision model fine-tuned from Qwen3.8-27B. The current KeyoAPI endpoint accepts text or image state with typed questions, and returns choice / noul probabilities in a single forward pass without generating free-form text. Use it for request routing, ticket triage and classification.
Call POST /v1/decisions with model cloudflare/clef and a Bearer API key. Check Model Square for current availability and pricing.
Context and image limits
The model context is 65536 tokens, but the text state limit on the serving path is much smaller: only approximately the first 2000 text state tokens are read. The remaining text is truncated and not read. Put decision-critical facts first; do not assume a 65536-token state will be processed. Images count separately from this text limit.
Attach at most 4 images as base64 data URLs in PNG, JPEG or WebP format. Remote image URLs are not supported. Each image must be below 300KiB on the current endpoint.
Answers and usage
The request contains model, state and questions. Each named question declares a supported type (choice or noul) and its instructions. For a choice question, criteria declares the candidate labels. The response contains answers and usage with input_tokens, output_tokens: 0 and cost. These are structured decision probabilities, not a generated chat message. The current KeyoAPI endpoint does not accept score questions or JSON-object state; send state as supported text or image content.
Starting price and actual settlement
Input starts at $0.063 per 1M tokens and output is $0. Settlement uses the actual request cost multiplied by 1.5. The catalog shows a starting rate, not a fixed rate for every request; actual input pricing may be higher when the serving route changes. Confirm current prices in Model Square before production traffic.
Quick start
Base URL: https://www.keyoapi.xyz/v1
Endpoint: POST /v1/decisions
Model: model=cloudflare/clef
curl https://www.keyoapi.xyz/v1/decisions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"cloudflare/clef","state":"A customer asks to reset password.","questions":{"route":{"type":"choice","instructions":"Route request","criteria":{"account":null,"billing":null,"technical":null}}}}'Auth uses the same Bearer API key as chat. Full notes: Keyo docs.
FAQ
Does Clef generate chat or free-form text?
No. A single forward pass returns typed choice / noul probabilities in answers. The current endpoint does not accept score questions. Use POST /v1/decisions.
Can Clef read a 65536-token text state?
Do not rely on it. 65536 is the model context length; the serving text state is truncated after approximately the first 2000 tokens and the remainder is not read. Images count separately.
Can I send remote image URLs?
No. Use up to 4 base64 data URL images in PNG, JPEG or WebP format, each below 300KiB.
Is $0.063/M the fixed input price?
No. It is the starting input rate. Output is $0, and each request settles at actual cost ×1.5. Actual input pricing may exceed the starting rate.