What is claude-model-compare?

This project is a local dashboard that shows how you personally use different Claude AI models inside Claude Code, the coding assistant tool. It works by reading transcript files that already exist on your own computer from past coding sessions, and turning them into charts and tables you can browse in a web browser. The idea is to answer questions you might have about your own habits, such as whether you use a more powerful model like Opus for different kinds of tasks than a lighter model like Sonnet, or whether one model tends to run longer sessions, call more tools, or hand off more work to helper agents. The dashboard shows things like session counts, how many messages and tool calls happen per turn, which tools each model uses most, and how work is spread across different categories over time. You can also download the underlying numbers as JSON or CSV files. Privacy is treated as a core design choice rather than just a promise. The tool only reads your transcripts, never writes to them, and only extracts numbers such as token counts, message counts, and timings. It never touches your actual prompts, responses, or code. Everything runs locally in your browser with no analytics and no network calls, and the generated data file is excluded from being uploaded if you were to share your own copy of the project. Getting started just requires Python 3.8 or newer, with no extra packages to install. You clone the repository, run one script to scan your Claude Code transcripts and build a data file, then run a second command to start a simple local web server and view the dashboard in your browser. The author is upfront that the dashboard describes patterns rather than judging model quality: differences between models mostly reflect which model you happened to choose for which task, not which model is better. There are 64 automated tests using fake, synthetic transcript data to check the tool's logic. The project is released under the MIT license, a permissive license that allows free use and modification, including commercially, as long as the copyright notice is kept.