Roboflow + OpenRouter
Run 300+ vision-language models from OpenRouter in Roboflow Workflows with one block, for OCR, captioning, classification, detection and structured answers.
How to connect OpenRouter to Roboflow
5 stepsAdd the OpenRouter block
Open a Workflow in the Workflow editor and add the OpenRouter block.
Choose a model
Paste an OpenRouter model slug, such as openai/gpt-4o-mini or qwen/qwen3.6-27b, into Model ID.
Pick a task and write a prompt
Select a task type, add your prompt, and add class names for classification or detection tasks.
Set billing and privacy
Leave the API key empty to bill your Roboflow credits, or paste your own OpenRouter key. Choose a privacy level, from no data collection to zero data retention.
Run and connect the output
Test the Workflow on an image, then send the answer or the parsed predictions to the next block.
What teams build with OpenRouter
Check a detection with a second opinion
Detect with RF-DETR, crop each region, and ask a VLM whether it is a real defect or a shadow.
Vote across models
Send the same crop to three models and take the majority answer.
Route by cost
Try a cheaper model first and escalate to a stronger one only when confidence is low.
Questions
No. By default the block uses a Roboflow-managed key and bills your Roboflow credits. Add your own OpenRouter key to bill OpenRouter directly.
Any OpenRouter model that accepts images. Paste its slug into Model ID.
Yes. In version 3 of the block, detection and classification answers come back as predictions that later blocks in the Workflow can use.
Why are companies choosing Roboflow?
The end-to-end platform for vision AI: create datasets, train models and deploy to production.
Flexible deployment modes
Move from a scalable cloud API to edge hardware like NVIDIA Jetson without managing complex dependencies.
Integrated workflow builder
Combine multiple AI models, custom business logic and third-party integrations into one production-ready pipeline in a low-code editor.
Model monitoring and alerts
Dashboards track confidence, inference time and request volume for your deployed models, and email alerts tell your team when confidence suddenly drops.
