If you use Claude Code a lot, you've probably seen this: the weekly limit bar creeping toward 100%, and no idea which model is eating it. Was it Opus on that big refactor? Sonnet in the background? Subagents?

Claude Code tells you how much of your weekly limit you've used. It doesn't tell you which models used it. So I built a small tool for that.

model-share pane refreshing

What it does

model-share is a Claude Code mod that adds a pane next to the chat:

  • Each model's share of your week. Everything you used this week counts as 100%, and each model gets its slice (mine right now: Fable 37%, Sonnet 34%, Opus 29%).
  • Your weekly limit, with pace. The bar turns green, amber or red depending on whether you're using it faster than the week is passing, and tells you when you'll hit 100% at this rate.
  • Follows your reset. The window matches your plan's weekly reset and starts over with it.
  • Auto refresh. Every 1, 3, 5 or 10 minutes, or off.

It works in the terminal, the Claude desktop app and the VS Code extension.

How it works

Claude Code already saves every session as a file under ~/.claude/projects/. Each reply in those files records which model wrote it and how many tokens it used.

model-share runs a small bundled Python script that:

  1. Reads this week's session files (all projects, subagents included)
  2. Counts each reply once (the same reply can appear more than once in a file)
  3. Adds up tokens per model, with cache reads weighted at 1/10, since they cost far less

The weekly window and limit percentage come from Claude Code itself.

A recount takes about 0.2 seconds, and because it only reads local files:

  • It uses zero tokens. No model calls.
  • Nothing leaves your machine. No network requests at all.

Install

In a Claude Code terminal session:

/plugin install model-share --marketplace shubhamverma8991/claude-model-share

Answer y, press Enter, and the pane opens. Type /model-share any time to bring it back, or /model-share every 5m to change the refresh interval.

Limitations (being honest)

  • Claude Code only. Chats on claude.ai or the mobile app aren't stored on your computer, so they aren't counted.
  • Shares are by tokens, not by limit. Claude Code doesn't report how much of the limit each model used, so this is an estimate. If a model counts more heavily against your limit per token, its true share is a bit higher.

What I learned building it

  • Claude Code's mod system lets you add real UI (panes, status lines, commands) without forking anything.
  • Session files can contain the same reply more than once, so deduplicating by message ID mattered a lot for accurate numbers.
  • Different clients get different data: the VS Code extension doesn't receive the limit percentage, so the mod remembers the last value any session saw.

Try it

It's free and MIT-licensed:
👉 github.com/shubhamverma8991/claude-model-share

If it's useful, a ⭐ on GitHub helps a lot. And tell me in the comments: what else would you want it to show? I'm thinking about daily breakdowns and week-over-week comparison next.