How to Check Qwen Code Usage and Limits

Qwen Code is Alibaba's terminal coding agent, built around the Qwen3-Coder models and open-sourced on GitHub. For a while it was one of the most generous free options around: sign in with a Qwen OAuth account and you got a pool of daily requests at no cost. That generosity is exactly what caught people out. The free quota was always small relative to a full day of agentic coding, it was tightened during 2026, and the free OAuth tier was ultimately discontinued. If you are still leaning on a constrained plan, the only way to pace yourself is to know — precisely — how much you have already burned.

How do I check my Qwen Code usage?

Qwen Code does not ship a rich historical usage dashboard. You can glance at a live counter inside a session, but it evaporates the moment you close the terminal, and it says nothing about which model ate your quota or what the equivalent spend would be. One command fills that gap:

$ npx whoburnedmore↳ scanning Qwen Code local logs…↳ parsed 612 turns across 41 sessions   QWEN CODE USAGE — last 14 days  ──────────────────────────────────────────  2026-06-17  requests 44   input 1,920,400  output 138,200  2026-06-16  requests 51   input 2,410,900  output 171,600  2026-06-15  requests 29   input 1,104,300  output  88,700   MODEL BREAKDOWN  qwen3-coder-plus     9.8M tokens   (71%)  qwen3-coder-flash    4.0M tokens   (29%)   14-day total:  13.8M tokens   est. $7.10

The request count per day is the line to watch. On the free OAuth tier your limit was denominated in requests, not tokens — so a day with 51 agentic turns can blow through the daily cap even if each turn was small. whoburnedmore surfaces that count next to the raw token totals, so you can see both ceilings at once.

What are Qwen Code's usage limits?

Qwen Code's ceilings have shifted with the plan landscape, so it helps to think in terms of the kind of limit rather than a single magic number:

1. 1

   Daily request quota (free OAuth tier)

   The original free tier handed you a fixed number of model requests per day. Each agentic turn — a plan step, a tool call, a follow-up — counts as a request, so the quota drained faster than newcomers expected. This is the tier that was reduced and then retired in 2026.
2. 2

   Per-minute / burst limits

   Even with quota left, rapid-fire requests can hit a short-window rate limit. Heavy tool-use loops are the usual trigger, because each loop fires a fresh request.
3. 3

   Token / context limits

   Separate from request counts, Qwen3-Coder models have a context window. Large repos stuffed into a single prompt burn tokens fast and can hit the per-request ceiling.

Because the headline number kept moving, treating Qwen's published cap as a moving target and your own measured usage as the fixed truth is the saner approach. The bar chart below is the kind of view that tells you whether you are pacing well or front-loading your whole quota into Monday 📊:

Jun 16 is the danger day — 51 requests against a tight cap leaves almost no headroom for the rest of the afternoon. Spotting that pattern is only possible with day-over-day history, which is precisely what the live in-session counter cannot give you.

How do I avoid hitting the Qwen quota?

Avoiding a mid-task lockout is mostly about spending requests deliberately. A few habits, each easier to keep once you can actually see your numbers:

1. 1

   Batch your asks

   Each agentic turn is a request. Bundling related changes into one well-scoped prompt spends fewer requests than a dribble of one-line follow-ups.
2. 2

   Pin a lighter model for routine work

   Route everyday edits to qwen3-coder-flash and save the heavier qwen3-coder-plus for genuinely hard reasoning. The model breakdown in whoburnedmore shows whether your split is sensible.
3. 3

   Trim the context you send

   Pointing the agent at a handful of relevant files instead of the whole repo keeps each request cheaper and less likely to trip a token ceiling.
4. 4

   Track the running total daily

   A quick npx whoburnedmoreat lunch tells you how much of the day's quota is gone, so the afternoon does not end in a surprise throttle.

How is Qwen Code cost estimated?

On the free tier the dollar figure is hypothetical — it is what your usage would cost at pay-as-you-go API rates. That number is still worth knowing: it tells you what a switch to a paid Qwen plan or the DashScope API would run you now that the free tier is gone. whoburnedmore prices input and output separately for each model and sums them:

Qwen3-Coder pricing is markedly cheaper than the frontier US models, which is a big part of its appeal — but output tokens still carry a premium over input, and a refactor-heavy week leans output-heavy. Seeing the real split is the only way to know whether a paid plan would actually be cheap for your pattern of work.

Free quota vs paid API vs whoburnedmore

Where does Qwen Code store its logs?

Qwen Code keeps session records in a hidden config directory in your home folder, written as it runs — each entry carrying a model identifier, a timestamp, and the token counts for that turn. whoburnedmore locates and parses these automatically, with no setup. If you moved the directory, point the tool at it explicitly:

$ npx whoburnedmore --qwen-dir ~/.qwen↳ using Qwen Code log dir: ~/.qwen↳ parsed 612 turns across 41 sessions   scan complete in 1.3s

Is it safe to point whoburnedmore at my Qwen logs?

Yes. whoburnedmore reads only the numeric and metadata fields — token counts, model names, request counts, timestamps. The prompt text and code context that Qwen Code sends to the model is never read or transmitted. Run npx whoburnedmore --dry-run to see the exact JSON that would be submitted, or npx whoburnedmore --local to keep the whole dashboard on your machine and skip the leaderboard entirely. 🛡️

And if Qwen Code is just one of several agents you run — alongside Claude Code, Codex CLI, or Gemini CLI — the same command lines them all up in one table. That cross-tool view is the whole point; see the complete AI token usage guide for the full picture.