Multimodal Decisions: Images in the Decisions API

​

Quickstart

IMAGE=$(base64 < listing.png | tr -d '\n')

curl https://openrouter.ai/api/alpha/decisions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d @- <<EOF
{
  "model": "cloudflare/clef-flash",
  "state": [
    "Listing title: Red square sticker",
    { "type": "image_url", "image_url": { "url": "data:image/png;base64,$IMAGE" } }
  ],
  "questions": {
    "matches_title": { "type": "noul", "instructions": "The photo shows the item in the title." },
    "color": {
      "type": "choice",
      "instructions": "What color is the item?",
      "criteria": { "red": null, "green": null, "blue": null }
    }
  }
}
EOF

{
  "model": "cloudflare/clef-flash",
  "answers": {
    "matches_title": { "type": "noul", "noul": 0.8607 },
    "color": {
      "type": "choice",
      "choice": "red",
      "probabilities": { "red": 0.9844, "green": 0.008, "blue": 0.0076 },
      "confidence": 0.9536
    }
  },
  "usage": { "input_tokens": 289, "output_tokens": 0, "cost": 0.00002601 },
  "id": "gen-dec-1791209504-cXRBLHSCBmsjQX93jkWe",
  "provider": "Cloudflare"
}

​

Image part format

{ "type": "image_url", "image_url": { "url": "data:image/jpeg;base64,/9j/4AAQ..." } }

• url must be a base64 data URL with the type image/png, image/jpeg, or image/webp. Remote http(s) URLs are not fetched.
• detail (auto, low, or high) is accepted, but Clef and Clef Flash ignore it.
• Put image parts directly in the top-level state array. Images nested inside other objects in state are not read as images.
• Send at most 4 images per request to Clef and Clef Flash, and at most 128 to GPT-6 Luna Decisions.
• Text items in the array are plain strings, not { "type": "text" } parts.

• A top-level images field. The API rejects it with a 400.
• The Responses API input_image part ({ "type": "input_image", "image_url": "data:..." }).
• A raw base64 string without the data:image/...;base64, prefix.

​

Limits

• Image size on Clef and Clef Flash. Workers AI estimates a request’s tokens from the base64 length before it processes the image, at about 4 base64 characters per token, and rejects the request with a 413 once that estimate passes its 65,536-token window. Keep each image under about 300 KB before encoding, and resize or recompress larger images. Billing uses the tokens the model actually processes.
• Text length on Clef and Clef Flash. Workers AI reads roughly the first 2,000 tokens of text in state and drops the rest without an error. Images are counted separately.
• See each model page for the current context length and pricing.

​

Related

• Jev Documentation covers the Decisions primitives (Choice, Noul, Score) in depth.
• Decisions API reference has the full request and response schema.