Computer vision

Read images and documents with vision models.

Send images, scans and video frames to Qwen3-VL 235B or Llama 3.2 90B Vision through one API, or train your own detector on a GPU instance.

L40Sfrom $1.26/hr

What you can build

Three jobs people run here. Each on a product you can start today.

Inference

Ask what is in a photo, chart or screenshot and get a reasoned answer, or structured JSON, from Qwen3-VL or Llama Vision.

Inference

Pull fields from invoices, receipts, forms and scans with PaddleOCR-VL or Qwen3-VL, with the layout understood, not only the text.

Instance

Train YOLO or fine-tune a segmentation model on your own labeled images on an L40S or A100 instance.

Examples

Connected by the line, Claude Code or Codex calls the Runcrate tools for you: launch, copy files, run, delete.

Hand this to your agent

Add a Runcrate model to this project: run curl -fsSL"https://runcrate.ai/agents.md?intent=build"and follow it.

Your agent asks for an API key. Create a key

Works inClaude CodeCodexOpenCodeClineRoo CodeKilo CodeCursor

example

Models and tools

Name	Role	Runs on
Qwen3-VL 235B	Visual reasoning and extraction	Inference API
Llama 3.2 90B Vision	Visual question answering	Inference API
Nemotron Nano 12B VL	Lightweight multimodal	Inference API
PaddleOCR-VL 0.9B	Document OCR	Inference API
Ultralytics YOLO	Custom object detection	Instance

How it works

01
Choose your approach
Call Qwen3-VL, Llama Vision or Nemotron through the Inference API, or launch an instance to train a model of your own.
02
Send images or frames
Pass images, screenshots, scans or frames from a video alongside a text prompt. The model reads them and answers.
03
Extract at scale
Process documents in bulk or add visual understanding to your product. The API bills per token, with no instance to run.

Pricing

GPUs are billed for the time they run: by the second on most GPUs, by the minute on a few.
A multi-node cluster for a month or longer: reserve one, and see every offer side by side.

Name

As billed