Vision (image input) Guide

You can pass images into messages on POST /v1/chat/completions and have them analyzed (OpenAI-compatible). There is no extra endpoint — an API key with the llm scope works as-is.

Which scope you need

api.aicu.ai API keys carry scopes (permission groups), and different endpoint families require different ones.

Scope	What it unlocks	Needed for Vision?
llm	/v1/chat/completions (text and image input alike)	✅
images	/v1/images/* (image generation)	No
tts	/v1/tts/* and /v1/audio/transcriptions (speech synthesis, transcription)	No

• Vision is just "pass an image to a chat completion," so the llm scope alone is enough. images is the scope for creating images
• Keys issued from the dashboard come with llm / tts / images by default
• If a company-issued key (least-privilege) returns 403, the scope is missing. Ask whoever issued it to add it

Calling it (curl, image URL)

curl -X POST https://api.aicu.ai/v1/chat/completions \
  -H "Authorization: Bearer $AICU_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Describe this image."},
        {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
      ]
    }],
    "max_tokens": 300
  }'

The trick is to make content an array instead of a string, and mix in parts of type: "image_url". messages is forwarded to the provider verbatim, so OpenAI's Vision spec works exactly as documented.

Calling it (Python, local image as base64)

import base64, os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["AICU_API_KEY"],
    base_url="https://api.aicu.ai/v1",
)

b64 = base64.b64encode(open("frame.jpg", "rb").read()).decode()
res = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "What's the highlight of this image? One line."},
            {"type": "image_url", "image_url": {
                "url": f"data:image/jpeg;base64,{b64}",
                "detail": "low",
            }},
        ],
    }],
    max_tokens=300,
)
print(res.choices[0].message.content)

• detail: low (low resolution, low cost) / high (tiled, high accuracy) / auto. Use low when analyzing a lot of frames

Supported models and pricing

Measured 2026-09-22 by sending the same small image to each model and asking what it shows.

Model	Images	Note
gpt-4o-mini	✅	Cheapest way to read an image
gpt-4o	✅
gpt-5.6-luna	✅	Lightest of the 5.6 family
gpt-5.6-terra	✅	Coding and image understanding
gpt-5.6-sol	✅	Same, with the deepest reasoning
gpt-6-astra	✅
kimi-k2.7-code	Preview	Declared multimodal, but it returned no answer in this test

The gpt-5.6 family reading images is what makes screenshot review from a coding agent (Cline, OpenCode) practical — see Coding agents.

• GET /v1/chat/models does not declare image support. There is no modalities or vision field, so tools that auto-detect capabilities assume text only. You have to tell them (OpenCode: declare modalities yourself — see Coding agents)
• Passing image_url to a text-only model such as deepseek-v3 returns an error
• Images are converted to tokens and billed (per OpenAI's rules; one image at detail: low ≈ 85 tokens. gpt-4o-mini counts about 33x the tokens but at a lower unit price, so the cost lands about the same — measured: one low image plus a short prompt ≈ 2,900 tokens). For per-model AP rates, see GET /v1/chat/models

Troubleshooting

Symptom	Cause
403 Forbidden	The key lacks the llm scope -> ask whoever issued it to add it
401 Unauthorized	Key is invalid or expired
Image fetch error	The URL is private or requires auth -> send it as base64 (data URL) instead

OpenAI's official docs (primary source for the compatible spec)

• Images and vision guide: https://platform.openai.com/docs/guides/images-vision
• Chat Completions API reference: https://platform.openai.com/docs/api-reference/chat/create

© 2026 AICU Inc.