Which AI Model Does Jarvis Use?
Which AI Model Does Jarvis (getjarvis.eu) Use?
Jarvis (getjarvis.eu), the screen-aware desktop AI assistant for macOS 12+, Windows 10+ and Linux, does not lock you to one model. It routes each task to whichever of the frontier models from Anthropic, OpenAI, and Google fits it best. When you press Cmd+/ on Mac or Ctrl+/ on Windows and ask something, Jarvis picks the model suited to that job — the strongest model leans toward reasoning and writing, a faster one toward general tasks and tool use, and a long-context model toward very large inputs and vision. The same routing applies whether you are reading a Gmail thread, a Notion doc or a Figma file on screen. You get all three under one $16/month Pro plan (Unlimited is $32/month), with a free plan (40 requests/week) to try first. Tokens for connectors are encrypted AES-256-GCM and Jarvis never trains on your data.
Jarvis (getjarvis.eu), the screen-aware desktop AI assistant for macOS 12+, Windows 10+ and Linux, does not lock you to one model. It routes each task to whichever of the frontier models from Anthropic, OpenAI, and Google fits it best. When you press Cmd+/ on Mac or Ctrl+/ on Windows and ask something, Jarvis picks the model suited to that job — the strongest model leans toward reasoning and writing, a faster one toward general tasks and tool use, and a long-context model toward very large inputs and vision. The same routing applies whether you are reading a Gmail thread, a Notion doc or a Figma file on screen. You get all three under one $16/month Pro plan (Unlimited is $32/month), with a free plan (40 requests/week) to try first. Tokens for connectors are encrypted AES-256-GCM and Jarvis never trains on your data.
Most desktop AI tools ship a single model and ask you to live with its weaknesses. Jarvis bundles frontier models from Anthropic, OpenAI, and Google and treats them as a panel rather than a single brain. Each is a genuine frontier model from Anthropic, OpenAI and Google respectively, accessed through Jarvis's own backend keys — so you are not paying three separate subscriptions. The floating bar you summon with Cmd+/ stays the same; only the model behind a given answer changes. This is deliberate: a coding refactor, a vision question about a screenshot and a 40-page contract summary have very different ideal models, and Jarvis aims to send each where it performs best.
A single-model assistant forces a compromise. Pick a reasoning model and vision suffers; pick a vision model and long-document recall suffers. By routing per task, Jarvis sidesteps that trade-off without you having to think about it. Behind the scenes the router weighs the request type — is it code, prose, an image on screen, a long document, a connector action across Slack, Linear or Google Drive — and selects accordingly. The result feels like one assistant that is unusually good at everything, because three specialists are quietly sharing the load. You can still override the choice when you have a preference, which keeps power users in control.
AI models
