Finding Vision Models on OpenRouter | Generated by AI

Home 2026.07

Question: How can I find image analysis (vision-language) models instead of image generation models on OpenRouter?

Answer:

OpenRouter does not separate them perfectly in the UI, so use these methods.

1. Search by model capability, not “image”

Avoid:

image model
image generation
flux
stable diffusion

Those are mostly generators.

Search:

vision
multimodal
image understanding
vision-language
VLM

Good VLM examples:

Model Type Use
google/gemini-2.5-flash VLM image + text → text
google/gemini-2.5-pro VLM complex image reasoning
anthropic/claude-sonnet-4 VLM document/image analysis
openai/gpt-5 VLM image understanding
qwen/qwen2.5-vl-72b-instruct VLM open vision model
qwen/qwen2.5-vl-32b-instruct VLM cheaper

2. Check OpenRouter model page

A vision model should show something like:

Input modalities:
  text
  image

Output modalities:
  text

You want:

image -> text

Not:

text -> image

Generation models usually show:

Input:
 text

Output:
 image

Examples:

black-forest-labs/flux.2-pro
stabilityai/stable-diffusion
openai/dall-e

These are NOT vision models.

3. Use OpenRouter API model list

You can query:

curl https://openrouter.ai/api/v1/models | jq .

Then filter:

curl https://openrouter.ai/api/v1/models \
 | jq '.data[] | select(.architecture.input_modalities | contains(["image"])) | .id'

You should see models like:

google/gemini-2.5-flash
anthropic/claude-sonnet-4
qwen/qwen2.5-vl-72b-instruct

4. For your ww vision-model test

I would test:

ww vision-model test \
  --model google/gemini-2.5-flash \
  --max-tokens 1000

or:

ww vision-model test \
  --model qwen/qwen2.5-vl-72b-instruct \
  --max-tokens 1000

Expected output:

Response:
The image is a blue square with a slight gradient...
finish_reason=stop

That finish_reason=stop is normal. The problem before was:

content=""

because FLUX has no text output.

For your use case (CLI agent screenshot understanding, ww vision-model), the ranking is probably:

  1. Gemini 2.5 Flash — cheap + fast
  2. Qwen2.5-VL — good OpenRouter option, China-friendly
  3. Claude Sonnet 4 — strong reasoning
  4. GPT-5 vision — strongest but expensive

FLUX should be moved to a separate command like:

ww image-generate test --model black-forest-labs/flux.2-pro

not vision-model.

Generated by AI. Curating and sharing still takes effort. If you find it useful, feel free to donate. WeChat: @lzwjavaWeChat QR · X: @lzwjava · Say hi 👋

Back Donate