Which Model Does Jarvis Use for Vision and Screenshots?

For vision tasks, reading your screen, interpreting a screenshot, parsing a chart or a Figma frame, Jarvis (getjarvis.eu) routes to a model with strong image understanding, typically a faster model, with a frontier model also capable of multimodal input. Because Jarvis is a screen-aware desktop assistant, vision is core: you press Cmd+/ on macOS or Ctrl+/ on Windows, and it can see whatever is on screen and answer about it. Ask it to extract text from an image, explain a diagram, or summarize a dashboard, and it picks a model that handles pixels well. Your screenshots are never used to train any model, stay encrypted in transit, and processing is GDPR-aligned on GDPR-aligned infrastructure. Jarvis runs on macOS 12+, Windows 10+, and Linux, starts with a free plan (40 requests/week), then $16/month, and connects to 30+ apps including Figma and Google Drive.

When you ask about something on screen, Jarvis captures the relevant view and sends it to a multimodal model. A frontier model is a natural fit for dense visual content, charts, tables, UI mockups, because of its strong image reasoning and large context. The model reads the image and grounds its answer in what's actually displayed: the numbers in a dashboard, the labels in a Figma frame, the error in a screenshot. You don't crop or upload manually; the floating bar handles capture, and routing handles model choice for the vision step.

Many real tasks mix seeing and doing. "Read this invoice screenshot and draft a reply in Gmail" needs vision to parse the image, then strong writing to compose the email, so Jarvis may use a faster model to read and the strongest available one to draft, within one flow. "Look at this chart and put the key numbers in Google Sheets" combines image reading with a connector action. Multi-model routing lets each step use the best engine, the vision model interprets pixels, the writing or reasoning model produces the output.

AI models