LLM Vitals
See how many tokens per second Claude Code and Codex are generating.
LLM Vitals shows the live tokens per second of your Claude Code and Codex sessions, with cache hit rate, context size and Codex rate limits. It reads the session logs these agents already keep on your computer, so it needs no API key and sends nothing anywhere. It works in the VS Code status bar, in Claude Code's own status line and as a live view in any terminal.
Live on the VS Code Marketplace, Open VSX and npmInstall from the Marketplace Open VSX, for Cursor and others npm: llm-vitals
What it does
Out of your way.
- Tokens per second of your latest Claude Code or Codex reply in the status bar, updated after every model call
- How long a request has been waiting while the agent is still working
- Cache hit rate, and context size (how many tokens the conversation takes up), with the share of the context window for Codex
- For Codex, how much of each rate limit is used
- A live table of every active session in the terminal, with npx llm-vitals
- One command, or one click in the editor, to show the speed in Claude Code's status line; one more to take it out
- Works with any model the agents run, read straight from their logs
Quick facts
Before you install.
- Price
- Free
- Works offline
- Yes, no connection needed
- Account
- None needed
- Platforms
- VS Code, Cursor, Antigravity and other VS Code based editors; any terminal with Node.js 20 or newer
- Status
- Out now
Questions
Asked and answered.
- Does LLM Vitals work offline?
- Yes, LLM Vitals works without an internet connection.
- Do I need an account to use LLM Vitals?
- No. There is no sign-up; just install it and start.
- Where can I get LLM Vitals?
- LLM Vitals is free on the VS Code Marketplace (https://marketplace.visualstudio.com/items?itemName=hamperpick.llm-vitals-vscode); on Open VSX, for Cursor, Windsurf and VSCodium (https://open-vsx.org/extension/hamperpick/llm-vitals-vscode); and on npm as llm-vitals (https://www.npmjs.com/package/llm-vitals).
- How do I see tokens per second in Claude Code?
- Install the LLM Vitals extension and the status bar shows it as soon as Claude Code replies. To see it inside Claude Code's own status line as well, click Enable when the extension asks, or run npx llm-vitals setup once in a terminal.
- Why is the speed lower than the model's advertised tokens per second?
- LLM Vitals divides a reply's output tokens by the time from the request being sent to the reply being logged. That time includes the wait before the first token, so it is the speed you actually get, a little lower than the model's raw generation speed.
- What does the context number in LLM Vitals mean?
- How many tokens the conversation takes up, that is everything the model read for its latest reply plus what it wrote. Codex logs its context window, so Codex sessions also show the share of it in use. Claude Code's logs don't record the window size, so its sessions show the token count only; in Claude Code's own status line, ctx 12% is the share Claude Code itself reports.
- Does LLM Vitals send my code or prompts anywhere?
- No. It makes no network requests. From each reply it uses only the token counts, model, times and project folder in the session logs on your computer, and never keeps or shows what you or the agent wrote.
- Can LLM Vitals show Codex speed inside the Codex terminal?
- No. Codex's status line only accepts its own built-in items, so Codex sessions appear in the editor's status bar and in the npx llm-vitals terminal view instead.
- Something else?
- Write to contact@hamperpick.in.