Claude Code Usage Limits Explained: What Changes on July 13, 2026 (and How to Check Yours)

Last updated: July 4, 2026 — tested on Claude Code v2.1.201, Max plan. This post will be updated after July 13 with measured before/after numbers.

If you use Claude Code on a Pro or Max plan, mark July 13 on your calendar: the temporary +50% weekly limit boost that’s been running since May 13 expires. And here’s the thing most posts get wrong — that’s the only thing expiring. The 5-hour limit doubling from May 6 has no announced end date.

This guide covers how the two-limit system actually works (with screenshots from my own /usage), what specifically changes on July 13, what drains your quota fastest, and your options when you hit the wall.

TL;DR: Two counters — a 5-hour rolling window and a weekly cap, shared across claude.ai, Desktop, CLI, and IDE. On July 13 (~6 PM PDT), only the weekly boost expires: expect to hit weekly walls roughly a third sooner from July 14. Check where you stand anytime with /usage.

The two limits: 5-hour window vs weekly cap

Claude Code runs on two overlapping counters, shared across every Claude surface — claude.ai chat, Claude Desktop, the CLI, and IDE extensions all draw from one pool.

Here’s my actual /usage (type it in any session):

Three things worth decoding in that screen:

• “Current session” is the 5-hour rolling window. The counter starts at your first prompt, not on a fixed clock — and the reset time is displayed in your local timezone (mine shows Asia/Seoul).
• “Current week (all models)” is the weekly cap. It resets at a fixed day and time assigned to your account — mine is Saturday 4am KST, every week, regardless of usage. The “everyone resets Monday” claim you’ll see in some posts is a myth.
• The second weekly bar varies by plan. Anthropic’s docs describe it as a “Sonnet only” pool on Max plans — on my Max plan (which includes Fable 5), it shows as “Current week (Fable)” instead. Either way, the pattern is the same: an all-models pool plus a separate pool for the tier that burns quota differently.

July 13, 2026: what actually expires (and what doesn’t)

Anthropic made two separate limit increases in May, and conflating them is the most common error in posts on this topic:

• May 6: 5-hour session limits doubled for Pro, Max, Team, and seat-based Enterprise. Peak-hours reductions removed for Pro/Max. No expiry announced.
• May 13: weekly limits +50%, announced as running “through July 13.”

So on July 13 (reported as 6 PM PDT — Anthropic’s announcement only says “through July 13”), only the weekly boost ends. Practical effect: your per-session experience shouldn’t change, but if you routinely brush against the weekly wall, expect to hit it roughly a third sooner starting July 14.

What happens after — full rollback to baseline or something in between — hasn’t been announced. I’ll update this post with my own before/after /usage readings once it lands.

Who’s affected: Pro, Max 5x, Max 20x, Team, seat-based Enterprise. API usage and usage-based Enterprise billing are separate.

Reading /usage like a pro

Scroll down in the /usage screen and there’s an attribution section: “What’s contributing to your limits usage?” — it breaks down recent consumption by skills, subagents, plugins, and MCP servers. Press d for the last 24 hours, w for the last 7 days.

One caveat printed right on the screen: it’s “approximate, based on local sessions on this machine — does not include other devices or claude.ai.” The progress bars reflect your account-wide state; the attribution below is local-only.

The same account-wide bars also live on the web at Settings → Usage:

What actually burns your quota

Ranked by impact, per Anthropic’s own docs:

1. Model choice. Official wording: Opus “uses meaningfully more of your quota” and “costs several times more per turn than Sonnet.” No exact multipliers are published — treat any precise ratio you read as a guess. Since Sonnet 5 became the default (June 30), the default is the cheap path.
2. Extended thinking. On by default; thinking tokens bill as output tokens, with budgets that can hit tens of thousands per request. Dial down with /effort or cap via MAX_THINKING_TOKENS.
3. MCP servers. Tool definitions are deferred by default now, but MCP still costs more context than plain CLI tools.
4. Agent teams — the real “7x.” That figure floating around applies to agent teams (each teammate is a separate Claude instance in plan mode), not regular subagents. Regular subagents are actually a saving technique — verbose output stays in the subagent’s context and only a summary returns.
5. Long conversations. Auto-compact consumes usage. Anthropic’s single most effective lever: /clear between unrelated tasks.

Hit the limit mid-task? Your options

1. Read the reset time — the block message shows exactly when you’re back, and the 5-hour window is at most a coffee break.
2. Downshift the model. /model to a lighter model — on Max, the premium-tier pool and the all-models pool drain at different rates, so switching models can un-block you immediately.

3. Turn on usage credits. The /usage screen shows the state (“Usage credits are off · /usage-credits to turn them on”). Enabled, prepaid funds bill at standard API rates after your included limits run out, with a monthly cap you control. Redemption is capped at $2,000/day.
4. Switch the session to API billing — Claude Code offers this explicitly; it never silently bills you. (Exception: a leftover ANTHROPIC_API_KEY env var can route you to API billing from the start — worth checking if your bills surprise you.)
5. Upgrade — Pro → Max 5x → Max 20x, prorated mid-cycle.

Pro vs Max: what the plans actually give you

Anthropic deliberately publishes no absolute token numbers — official language is “Max 5x provides 5 times more usage per session than Pro.” The last concrete estimates date to August 2025 (Pro ≈ 40–80 hours of Sonnet per week; Max 5x ≈ 140–280h; Max 20x ≈ 240–480h) — Sonnet-4-era numbers, before the May 2026 changes, so treat them as a stale baseline only.

I’m tracking a full week of my own /usage on a Max plan right now and publishing the day-by-day numbers — including whether someone with my usage pattern actually needs Max at all — in an upcoming Pro vs Max breakdown.

One open bug to know about (Max plans)

Issue #57050: users report usage of the non-premium model draining both weekly pools simultaneously. It’s labeled duplicate and remains unresolved. If your all-models bar climbs on days you only used the cheaper tier, you’re not imagining it.

FAQ

When exactly does my weekly limit reset? At a fixed day+time assigned to your account, shown in your local timezone in /usage (and at claude.ai Settings → Usage). It doesn’t shift based on your activity.

Does claude.ai chat usage count against Claude Code? Yes — one shared pool across claude.ai, Desktop, CLI, and IDE.

Does the July 13 change affect the 5-hour limit? No. Only the +50% weekly boost expires. The May 6 doubling of 5-hour limits has no announced end date.

Can I see what’s eating my quota? /usage → scroll to the attribution section → d/w to toggle the window. Local sessions only, but that’s usually where the surprises are (MCP servers and subagent-heavy workflows).

Can I buy my way past a limit? Yes — usage credits at API rates (/usage-credits), or switch the session to API billing.

Last updated July 4, 2026. This post will be updated on July 14 with measured before/after numbers once the weekly boost expires.

WorkflowDen is an independent site, not affiliated with or endorsed by Anthropic, PBC. Screenshots are from my own /usage on Claude Code v2.1.201.