Vision Models

Send images alongside text prompts

Vision-capable models accept images alongside text using the standard OpenAI content-parts format. The gateway routes your request to the model as-is — no re-encoding, no extra fees beyond the model's normal input token price for the image content.

Sending an image

Pass an array of content parts instead of a plain string. Images can be public URLs or base64 data URLs.

python
resp = client.chat.completions.create(
    model="claude-opus-5-5",
    max_tokens=500,
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "What's in this screenshot? Be specific."},
            {"type": "image_url", "image_url": {"url": "https://example.com/ui.png"}},
        ],
    }],
)

Base64-encoding a local file

python
import base64

with open("chart.png", "rb") as f:
    b64 = base64.b64encode(f.read()).decode()

part = {"type": "image_url",
        "image_url": {"url": f"data:image/png;base64,{b64}"}}

Finding vision models

The catalog marks every model's input modalities. Filter oninput_modalities containing "image":

bash
curl -s https://tokenhub-staging.coality.io/api/public/models | jq '[.data[] | select(.input_modalities | index("image")) | .id]'

Most frontier models in the catalog accept images, including claude-opus-5-5, gpt-6-astra, gemini-4-argon, and qwen3-8-max. Check a model's detail page for its exact modality list.

Practical tips