How to Check Gemini CLI Usage and Cost

Google launched Gemini CLI as a free, open-source terminal agent in mid-2025. It quickly attracted developers who wanted a generous free tier alongside Claude Code and Codex CLI. But there is a notable gap in the tooling: while Gemini CLI tracks usage internally to enforce rate limits, it does not expose a spend view to the user. You cannot type a command and see how many tokens you burned today or what the equivalent dollar cost would be at paid-tier rates. That gap is what whoburnedmore closes.

Why doesn't Gemini CLI show my token usage?

The short answer: it was not a priority at launch. The free tier is metered by requests per minute and per day rather than tokens, so the UI focused on request counts. Token spend only matters once you leave the free tier and start billing by the token — and by then Google expects you to use the Cloud Console billing dashboard or the AI Studio usage page.

The problem is that neither the Cloud Console nor AI Studio know about usage from the CLI unless you are authenticated with an API key. If you are using Gemini CLI on the free tier with OAuth login, your usage is invisible in every Google dashboard. The feature request tracking a native /usage command for Gemini CLI has been open since late 2025.

Where does Gemini CLI store its session logs?

Despite the missing UI, Gemini CLI does write structured logs to disk. By default, they live under ~/.gemini/sessions/ on macOS and Linux. Each session is a JSONL file with one record per turn, including the model name, token counts, and timestamp. whoburnedmore reads these to reconstruct your usage history:

$ ls ~/.gemini/sessions/2026-06-14_session_a1b2c3.jsonl2026-06-13_session_d4e5f6.jsonl2026-06-12_session_g7h8i9.jsonl $ head -1 ~/.gemini/sessions/2026-06-14_session_a1b2c3.jsonl | python3 -m json.tool | grep -E 'model|token'    "model": "gemini-2.0-flash",    "inputTokenCount": 24381,    "outputTokenCount": 1847,

What counts as a Gemini CLI token?

Google uses the same transformer-style tokenization for Gemini models as other providers. Roughly speaking, one token maps to three or four characters of English text. Code is denser — a 200-line Python file typically encodes to 1,500–2,500 tokens depending on naming conventions and whitespace. Gemini 2.0's context window is 1M tokens, which sounds enormous until you realize a large codebase passed repeatedly to the model can fill it within a handful of sessions.

What does Gemini CLI actually cost?

On the free tier, cost in dollars is zero — up to the rate limits. Beyond that, billing switches to pay-per-token. whoburnedmore computes both the token total and what that usage would cost at the current Gemini API rates, so you can monitor whether you are approaching the threshold where billing kicks in or — if you are already on a paid plan — how your bills are accruing:

At mid-2026 rates: Gemini 2.0 Flash is around $0.075 per million input tokens and $0.30 per million output tokens — inexpensive for individual requests, but sessions that repeatedly send large context windows accumulate quickly. Gemini Pro is priced roughly 10x higher. A month where 20% of your requests were quietly routed to Pro instead of Flash can increase your bill by 2–3x.

$ npx whoburnedmore↳ scanning ~/.gemini/sessions/…↳ found 412 entries across 31 sessions   GEMINI CLI USAGE — last 30 days  ─────────────────────────────────────────────  2026-06-14   Flash  1,841,200   Pro    287,400  2026-06-13   Flash  2,304,700   Pro     94,100  2026-06-12   Flash    987,300   Pro    341,900   MODEL BREAKDOWN  gemini-2.0-flash       28.1M tokens   $2.11  gemini-2.0-pro          4.0M tokens   $3.98   30-day total:  32.1M tokens   est. $6.09

Note June 12: Pro tokens were 35% of that day's total, which drove a disproportionate share of the estimated cost. Gemini CLI routed those requests automatically — whoburnedmore makes the split visible so you can decide whether to configure a model preference in your ~/.gemini/config.json.

Which Gemini CLI free-tier limits should I watch?

The free tier enforces two independent ceilings. Hitting either one pauses your requests until the window resets. Neither is visible in the CLI itself — you find out when requests start returning 429 RESOURCE_EXHAUSTED:

Running whoburnedmore in local mode

1. 1

   Standard run (leaderboard)

   npx whoburnedmore — scans all supported tools, prints the breakdown, and optionally submits your totals to the public leaderboard.
2. 2

   Local-only mode

   npx whoburnedmore --local — opens a private browser dashboard with the same charts, nothing submitted. Useful if you want the visual without sharing.
3. 3

   Dry run to inspect what is shared

   npx whoburnedmore --dry-run — prints the exact JSON payload that would be sent, so you can verify only aggregate totals are included, not prompts or file names. 🛡️