autoloops
← Models
Model APIJev

Gemma 4 31B Jev

Calibrated answers · Text and images

Ask questions about a conversation or a photo and get a probability for each answer. Questions can be yes/no, multiple choice or a score on a scale, and one call can ask many of them.

Model details

EndpointPOST /v1/systemone
Question typesYes/no, multiple choice, score
InputChat messages, optionally with JPEG or PNG images
OutputA probability for each option. The model generates no text.
Questions per callMany. The model reads the conversation once and scores every question against it.
CompatibilityText-only requests match TypeSafe's Jev API

On Autoloops

HardwareB200
APIJev (POST /v1/systemone)
RegionUS West
Pricing$0.20 input, $0.14 cached per 1M tokens. No output tokens.
Model IDgemma-4-31b-it

Capabilities

  • Returns calibrated probabilities for yes/no, multiple-choice and score questions
  • Answers questions about photos, such as what is in them and what state things are in
  • Scores many questions in one request. Repeat requests on the same conversation hit the cache.
  • Returns JSON with one entry per question, so there is no free text to parse
  • Bills input tokens only, with cached tokens at the cached rate

Ask questions about an image

IMG=$(base64 < photo.jpg | tr -d '\n')    # JPEG or PNG

curl https://api.autoloops.ai/v1/systemone \
  -H "Authorization: Bearer $AUTOLOOPS_API_KEY" \
  -H "Content-Type: application/json" \
  -d @- <<EOF
{
  "model": "gemma-4-31b-it",
  "state": [
    {"role": "user", "content": [
      {"type": "image_url", "image_url": {"url": "data:image/jpeg;base64,$IMG"}}
    ]}
  ],
  "questions": {
    "animal": {"type": "choice", "instructions": "Which animal is in the photo?",
      "criteria": {"dog": null, "cat": null, "none": null}},
    "on_sofa": {"type": "noul", "instructions": "Is the animal on the sofa?"}
  }
}
EOF

# → {"answers": {"animal": {"type": "choice", "choice": "dog", "probabilities": {"dog": 0.94, …}},
#                "on_sofa": {"type": "noul", "noul": 0.98}}, "usage": {…}}

More models