Claude Code Usage Limits
A practical guide to Claude Code usage limits, model choice, context window management, subscription vs API metering, and ways to avoid wasted quota.
Claude Code usage limits are the amount of work your account can run before a reset window, credit purchase, upgrade, or pay-as-you-go billing decision is needed. They are not only message counts. Model choice, context size, file reads, tool calls, and session length all affect how quickly usage is consumed.
Last checked on May 24, 2026. Exact allowances, reset windows, model names, and credit options change. Trust the message shown inside Claude Code and your account dashboard over any third-party table.
Short Answer
If you sign in with a Claude subscription, your usage is governed by the plan and shared with other Claude surfaces. If you sign in with an API key or provider account, usage is billed by token and provider rules. Either way, long sessions with large context are the fastest way to burn through limits.
What Consumes Usage
Every meaningful turn can include:
- The conversation so far.
- Project context such as
CLAUDE.md, files read, rules, skills, and MCP outputs. - Your new prompt.
- Tool calls, file diffs, search results, and command output.
- Model-specific reasoning effort.
The hidden multiplier is accumulated context. A session that already read many files and produced many diffs sends much more context on later turns than a clean session.
Subscription vs API Limits
| Login path | What running out looks like | What to check |
|---|---|---|
Pro, Max, Team, or Enterprise via /login | Claude Code shows a limit or reset message. | /status, /model, usage messages, plan settings, usage credit options. |
| API key from Claude Console | There may be no hard stop; the account is charged for usage. | /status, /cost or /usage if available, Console usage, spend limits. |
| Bedrock, Vertex AI, Microsoft Foundry | Provider billing and provider rate limits apply. | Cloud dashboard, provider quotas, Claude Code authentication settings. |
| Enterprise seat | Usually governed by an organization pool or policy. | Admin analytics, org limits, reset message, policy configuration. |
Commands That Help
| Command | Use it for |
|---|---|
/model | See available models and switch between Sonnet, Opus, Haiku, or account-specific options. |
/status | Confirm which account and credential path the session is using. |
/clear | Start a fresh conversation while keeping project memory available. Best between unrelated tasks. |
/compact | Summarize the current conversation to free context while continuing the same task. |
/context | Inspect what is loaded into context and find sources of bloat. |
/cost or /usage | Check session spend or token usage when the command is available for your billing path. |
Five Habits That Stretch Usage
- Clear between tasks. If the next prompt would make sense in a brand-new terminal, run
/clear. - Use Sonnet by default, Opus for hard planning, and Haiku for fast simple work when available.
- Ask for a plan before multi-file changes.
- Reference paths instead of pasting whole files.
- Keep
CLAUDE.mdunder control. Helpful memory saves turns; stale memory wastes context.
When You Hit a Limit
- If Claude Code shows a reset time, decide whether to wait, switch to a lighter model, buy usage credits if your plan allows it, or move the sprint to API billing.
- If a context window is full, use
/compactor/clear. That is different from a plan usage limit. - If API spend is unexpectedly high, stop the script first, then check whether the session was carrying a large transcript, large MCP output, repeated retries, or the wrong model.
- If limits suddenly feel different, check official release notes, your active model, and whether a new feature such as extended context changed usage behavior.