Reverse engineering with Google Gemini
This page lists projects where Google's Gemini models did part of the reverse engineering. Gemini appears less often in the directory than Claude or Codex, but the entries show it working on quite different targets: the raster format a printer driver sends, a hidden softkey API inside an Android phone's system framework, and an academic user study that compared several models. Gemini can run inside a general agent such as OpenCode, as one project here did, or Gemini CLI, and some authors simply paste decompiled code into a chat. One author reported a run that cost a few cents in tokens, which makes cheaper Gemini models worth trying for narrow, well defined tasks such as reading decompiled Smali. As with any model, the person still tests the result on the real device or program. Each entry links the original writeup or repo and lists the model version, tools and outcome. Submit your own Gemini project to help fill this page.
- Projects
- 3
- Outcomes
- Works2Research1
- Models used
- Gemini3Claude1GPT1
The projects
Share the writeup or repo, and what the AI got right and wrong.
Related guides
Common questions
Yes. Projects here used Gemini to work out a printer's raster format and to recover a hidden Android framework API.
General agents such as OpenCode and Gemini CLI can run Gemini with local tools. MCP servers such as REA list Gemini CLI among supported clients.
Fewer public writeups name Gemini so far. The page fills as people submit projects.
