How to Check Kimi CLI Usage and Cost

Kimi CLI is Moonshot AI's terminal coding agent, driven by the Kimi K2 family of models and wired for MCP servers and reusable skills. It is genuinely pleasant to drive from the prompt, but it is built like a tool, not like a billing console: it prints the token count for the turn it just finished, and that number evaporates the instant the session ends. There is no dashboard, no “this month” tab, and nothing that tells you whether last week's big agent loop quietly cost more than you expected.

How do I see my Kimi CLI token usage?

Kimi writes a structured log of each agent run to your machine. You do not have to parse it by hand — one command sweeps those logs and prints a real breakdown:

$ npx whoburnedmore↳ detecting installed coding agents…↳ kimi-cli logs found · 19 runs parsed   KIMI CLI USAGE  ────────────────────────────────────────  2026-06-16  in 1,902,400  out 154,300  2026-06-15  in 2,640,100  out 221,800  2026-06-14  in   880,950  out  61,200   BY MODEL  kimi-k2          11.4M tokens  kimi-k2-turbo     3.2M tokens   this month:  14.6M tokens   est. cost shown inline

Because whoburnedmore lists Kimi among the agents it understands, you do not configure anything: it discovers the log directory, walks each recorded run, and folds the per-turn token fields into day-level and model-level totals. The K2 turbo variant and the standard K2 model are reported separately, since their per-token rates differ.

What Kimi's built-in readout gives you

Inside an active Kimi session you can glance at the tokens for the current exchange, and that is useful for catching a context window that is ballooning mid-task. What it cannot do is look backwards. Three gaps are worth naming:

1. 1

   No persistence across sessions

   The live token figure belongs to the running process. Quit Kimi and the count resets to zero; yesterday's agent marathon leaves no on-screen trace.
2. 2

   No per-day or per-model rollup

   You see one run at a time, never “Tuesday burned 4M tokens” or “K2-turbo was 22% of my spend.” The aggregation simply isn't built in.
3. 3

   No cost translation

   Kimi reports tokens, not dollars. Turning K2 token counts into an estimated bill is left entirely to you — which is exactly the chore whoburnedmore automates.

What does Kimi CLI cost?

Moonshot bills the Kimi API by tokens, charging input and output at different rates, with the turbo tier priced apart from the standard model. whoburnedmore does not hard-code those numbers — see Moonshot's pricing for the current figures — but it applies the right rate to each model bucket and sums them so you get a single estimate:

The split matters more than people expect: output tokens are the pricier side, so a few verbose agent runs can dominate the bill even when input dwarfs output by raw count. The donut below shows how a typical month divides across the K2 models and the system prompts the agent keeps replaying:

How do I track Kimi usage over time?

Tracking trend is the whole point, and it is the one thing the in-session readout can never give you. Because whoburnedmore reads the saved run logs, it can rebuild history you never explicitly recorded — yesterday, last week, the whole month — in three steps:

1. 1

   Run it on a cadence

   Drop npx whoburnedmoreinto your shell once a day, or just run it whenever you're curious. Each run re-reads the full log, so you never miss a session.
2. 2

   Read the day-by-day column

   The daily table makes spikes obvious. A day of heavy MCP tool-calling, where the agent re-sends context on every loop, lights up immediately against quieter days.
3. 3

   Compare model mix month over month

   Watch the K2 vs K2-turbo ratio shift as your habits change. Leaning on turbo for speed shows up as a cost line you can actually see and decide about.

Is reading my Kimi logs private?

Your prompts, your file names, and your code never travel anywhere. whoburnedmore lifts only the numeric fields — token counts, model labels, timestamps — and the single thing that can leave your machine is the daily aggregate, and only if you opt to post it to the leaderboard. Want to be sure? Run npx whoburnedmore --dry-run to print exactly what would be submitted, or npx whoburnedmore --local to open a dashboard that never leaves your laptop. 🛡️

Kimi CLI gives you a fast, MCP-aware agent and a glimpse of the current run. whoburnedmore supplies the part Moonshot left out of the terminal: a durable, day-and-model view of what you actually burned — and a leaderboard spot to prove it.