Claude Code Usage Limits: Check Them, Then Stretch Them
How Claude Code usage limits work on Pro and Max, how to check yours with /usage in ten seconds, and the token-diet habits that make any plan last longer.
Evgenii Arsentev · MD, PhDReviewed for accuracy by Evgenii Arsentev, MD, PhD · 2026-06-12
Quick orientation: Claude Code usage limits depend on your plan — Pro has a base allowance, Max plans give you 5x or 20x more usage than Pro, and the /usage command inside Claude Code shows exactly where you stand right now. There's no public table of 'N requests per day', because consumption is measured in tokens and varies wildly with how you work — which, as we'll see, is actually good news, because it means your habits control your limits.
I build with Claude Code every day, often several projects in parallel, and I used to hit limits weekly until I understood what actually eats the allowance. Spoiler from Anthropic's own cost docs: token costs scale with context size — the more conversation history and files Claude has to re-read on every message, the faster your meter runs. Most people who complain about limits are paying a stale-context tax without knowing it.
What are the Claude Code usage limits on each plan?
The ladder: the free claude.ai plan has the simplest limit of all — zero, it doesn't include Claude Code. Pro ($20/month, $17 annual) includes Claude Code with a base usage allowance. Max (from $100/month) is literally sold as a limits upgrade: you choose 5x or 20x more usage than Pro, plus higher output limits for all tasks. Team standard seats include Claude Code with more usage than Pro, and premium seats carry 5x a standard seat. If you instead pay per token through a Console (API) account, there's no allowance at all — just a bill, and optional workspace spend limits an admin can set.
Why is there no fixed 'requests per day' number?
A token, for the non-programmers this site is for: the unit AI text is measured in, roughly three-quarters of a word. Everything you send AND everything Claude reads — your files, its own earlier answers, tool outputs — gets counted. That's why there's no public 'N requests per day' table, and why two people on identical Pro plans can have wildly different experiences of 'the limit'.
How do I check my Claude Code usage limit?
Inside any session, type /usage. On Pro, Max, Team and Enterprise plans it shows your plan usage bars, activity stats, and — my favorite part — a breakdown of what's consuming your allowance, attributed to skills, subagents, plugins and individual MCP servers as percentages. Press d or w to flip between the last 24 hours and the last 7 days. The figures are approximate and built from local session history on this machine, so usage from your other devices or from chatting on claude.ai isn't included.
$ claude > /usage Plan usage: ▓▓▓▓▓▓░░░░ bars + activity stats Breakdown: which skills, agents and MCP servers ate what press d / w — last 24 hours vs last 7 days
How do I stop surprise spending?
Two more tools worth knowing. /context shows what's sitting in your context window right now — the live view of what every message re-processes. And on Pro and Max, /usage-credits lets you set a monthly spend cap on extra usage credits; if you hit the cap while credits remain, Claude Code prompts you to raise or remove it without leaving the terminal. That's the 'how do I stop surprise spending' answer in one command.
Why do I hit limits so fast — and what eats them?
Almost always one of four leaks. One: never clearing context — a conversation that's been running since breakfast re-processes its whole history with every message. Two: running everything on the biggest model when Sonnet handles most everyday building for less than Opus. Three: extended thinking on hard mode for trivial tasks — thinking tokens are billed as output tokens, and the default budget can run to tens of thousands of tokens per request. Four: a bloated setup — an overstuffed CLAUDE.md (the docs suggest keeping it under 200 lines) and unused MCP servers quietly taxing every session. None of these feel like waste while you're doing them. All of them show up in /usage.
How do I stretch my limits without upgrading?
The token diet — straight from Anthropic's cost docs
- 1Use /clear every time you switch to an unrelated task. Stale context wastes tokens on every subsequent message — this single habit is half the battle.
- 2When a long session matters, run /compact with instructions, like: /compact Focus on code samples and API usage — Claude summarizes history and keeps what you named.
- 3Switch models with /model: Sonnet for everyday work, Opus only for genuinely complex reasoning; for simple subagent jobs there's even Haiku.
- 4Run /mcp and disable servers you're not actively using; run /context to see what's occupying space.
- 5For big tasks, press Shift+Tab into plan mode first — approving an approach before code is written prevents the most expensive thing in AI building: confidently wrong work you pay to redo.
- 6Course-correct early: Escape stops Claude mid-flight, /rewind restores a checkpoint. Ten wasted minutes is cheaper than a wasted hour.
- 7Write specific prompts. 'Add validation to the login form in auth.js' touches a few files; 'improve this codebase' triggers a broad scan you'll pay for.
Claude Code spends a trickle of tokens on background work — summarizing past conversations so /resume works, status checks for some commands. It's tiny (typically under $0.04 per session), but if you leave agent teams or extra sessions running idle, each keeps its own context alive. Close what you're not using.
Lab: a one-week limit audit
0/5Here's the reframe I want you to leave with: limits aren't a wall, they're a fuel gauge. You wouldn't drive everywhere in first gear and blame the tank. Run the one-week audit before you pay for Max — and if you still bump the ceiling after the diet, congratulations: you're genuinely building enough to deserve the upgrade.
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →References
Selected peer-reviewed papers and preprints on the AI research behind this topic.
- [1]Zhai (1997). Exploiting Context to Identify Lexical Atoms -- A Statistical View of Linguistic Context. arXiv:cmp-lg/9701001
- [2]Liu et al. (2025). Thus Spake Long-Context Large Language Model. arXiv:2502.17129
- [3]Fu et al. (2024). Data Engineering for Scaling Language Models to 128K Context. arXiv:2402.10171
- [4]Anderson et al. (2018). An Ontology-Based Reasoning Framework for Context-Aware Applications. arXiv:1805.09012
- [5]Guo et al. (2025). Long Context Tuning for Video Generation. arXiv:2503.10589
- [6]Liu et al. (2025). Steering Multimodal Large Language Models Decoding for Context-Aware Safety. arXiv:2509.19212
- [7]Cífka et al. (2022). Black-box language model explanation by context length probing. arXiv:2212.14815
Author
Evgenii Arsentev
MD, PhD · AI transformation executive
Reading is the blue pill
Want to actually build this?
Guides explain. The free course transforms — personalized, gamified, and built to get you shipping fast.
Start the free course →