Which Model Does Jarvis Use for Coding Tasks?

For coding, Jarvis (getjarvis.eu) routes to a frontier model, which is its strongest model for code generation, refactoring, and reviewing diffs. When you point Jarvis at code on your screen, in an editor, a GitHub PR, or a Linear ticket, it reads what's visible, reasons about it with Claude, and proposes changes or explanations. You trigger it with Cmd+/ on macOS or Ctrl+/ on Windows. If a task involves a very large codebase dump or long logs, Jarvis may lean on a long-context model, or use a faster model for quick scripting. You can force any of them. Jarvis runs on macOS 12+, Windows 10+, and Linux, connects to GitHub among 30+ apps, encrypts OAuth tokens with AES-256-GCM, is GDPR-aligned, and never trains on your code. It starts with a free plan (40 requests/week), then $16/month.

Code quality depends on careful reasoning about edge cases, types, and intent, and a frontier model is consistently strong there. Jarvis routes most programming work to it: explaining a confusing function, writing a fix, reviewing a GitHub diff, or translating a Linear bug report into a patch. Because Jarvis sees your actual screen, Claude reasons about the real code in front of you, not a description. That cuts the back-and-forth, you don't have to paste everything, and the model works from the same context you're looking at.

Two cases shift the routing. First, huge inputs, a long stack trace, a sprawling file, or many files at once, where a frontier model's large context window lets it hold everything. Second, fast throwaway scripting or boilerplate, where a faster model's speed is convenient. You can force either: "use Gemini for this whole repo" or "a faster model, write a quick bash one-liner." Multi-model routing means Jarvis isn't boxed into one engine's strengths; it matches the model to whether you need depth, breadth of context, or speed.

AI models