Claude Code Usage Limits (2026): 5-Hour Caps Doubled May 6, Weekly Limits by Plan

What Changed on May 6, 2026

Most pages about Claude usage limits were written before May 2026 and are now wrong. On May 6, 2026, Anthropic announced three changes, effective the same day:

• Claude Code's five-hour rate limits doubled for Pro, Max, Team, and seat-based Enterprise plans.
• The peak-hours limit reduction was removed for Pro and Max Claude Code accounts. Limits no longer shrink during high-demand hours.
• Opus API rate limits were raised for Console accounts.

The capacity came from a compute agreement with SpaceX covering the Colossus 1 data center: more than 300 megawatts, over 220,000 NVIDIA GPUs, online within the month of the announcement. Anthropic stated the capacity would directly improve availability for Pro and Max subscribers.

Limits by Plan: Pro, Max 5x, Max 20x, Team

Claude Code requires a Pro, Max, Team, Enterprise, or Console account. Anthropic's setup docs state it plainly: the free Claude.ai plan does not include Claude Code access.

Anthropic stopped publishing fixed prompts-per-window and hours-per-week figures. What it publishes instead is relative capacity: Max 5x provides 5 times more usage per session than Pro, Max 20x provides 20 times more. Those multipliers, the May 6 doubling, and the plan prices are the concrete numbers that exist.

Enterprise pricing is seat price plus usage billed at API rates. Pricing details for every tier are on Claude Code pricing and Claude Code subscription plans.

How the 5-Hour Session Works

Usage is metered against your plan's five-hour session limit. The session opens with your first prompt and covers everything you do for the next five hours, in Claude Code and in Claude chat, since both draw from the same pool.

Worked example: you send your first prompt at 9:00 a.m. That opens a session running until 2:00 p.m. Every prompt, file read, tool call, and thinking block until then counts against that session's allowance. Burn through it by 10:30 a.m. and subscription usage pauses until the session resets at 2:00 p.m. Send nothing until 3:00 p.m. and your next prompt opens a fresh session with a full allowance.

Before May 6, 2026, Pro and Max accounts also had reduced limits during peak hours. That reduction is gone: your five-hour allowance is the same at 2 p.m. on a Tuesday as at 2 a.m. on a Sunday.

Weekly Limits

Weekly limits sit on top of the session limit and accrue across sessions. Hitting a weekly cap locks usage until its reset even if your current five-hour session has allowance left.

• Max plans carry two weekly limits: one that applies across all models, and another for Sonnet models only.
• Resets are visible per limit. The usage page shows when each weekly limit resets, tracked separately by model group.
• Weekly limits are why bursts feel fine and sustained use does not. A heavy weekend can leave you session-capped on Monday with weekly headroom, or weekly-capped with sessions to spare.

Anthropic's support docs no longer state fixed hours-of-Sonnet or hours-of-Opus per week. Treat any page quoting exact weekly-hour figures as stale; the published facts are the structure (session + two weekly caps on Max) and the per-plan multipliers.

What Burns Your Quota

Your prompts and Claude's visible responses are a minority of token flow. Cognition measured coding agents spending 60% of their time on search. The rest of the burn is configuration you control:

For scripted runs, claude -p --bare skips hooks, skills, plugins, MCP, and CLAUDE.md discovery entirely, which strips that fixed overhead from non-interactive calls.

How to Check Your Usage

The dashboard is the source of truth for reset times because it tracks each weekly limit separately. Check it before a long session rather than after the lockout message.

Detect Limit Hits Programmatically

Claude Code's hook system has a dedicated event for turns that end in API errors: StopFailure, with matchers including rate_limit, overloaded, and billing_error. A hook on rate_limit turns a silent lockout into a notification or an automated fallback:

{
  "hooks": {
    "StopFailure": [
      {
        "matcher": "rate_limit",
        "hooks": [
          {
            "type": "command",
            "command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/on-rate-limit.sh"
          }
        ]
      }
    ]
  }
}

The script receives session JSON on stdin (session_id, transcript path, cwd) and can post to Slack, log the timestamp to correlate with your session windows, or queue the task for the reset. Full event reference in our Claude Code hooks guide; settings file mechanics in settings.json explained.

The Agent SDK Credit (June 15, 2026)

Starting June 15, 2026, non-interactive usage on subscription plans stops competing with your interactive sessions. Agent SDK projects, claude -p, the GitHub Actions integration, and third-party apps authenticating with your subscription draw from a separate monthly credit:

When the credit is exhausted, usage flows to usage credits at standard API rates if you have them enabled; otherwise requests stop. Unused credit does not roll over. Practical consequence: CI jobs and cron agents no longer eat the five-hour window you need for interactive work, but they now have their own hard monthly budget.

When You Hit the Wall: API Rates

A Console (API) account has no five-hour sessions and no weekly caps. You pay per token:

Cache reads bill at 0.1x the input price, which matters because Claude Code's fixed overhead (system prompt, tool definitions, CLAUDE.md) is exactly the kind of repeated prefix that caching absorbs. Sonnet 4.6 includes the full 1M-token context window at standard pricing, with no surcharge tier above 200K.

Arithmetic for scale: a session that processes 2M input tokens and generates 200K output tokens on Sonnet 4.6 costs $6 + $3 = $9 at uncached rates, less with cache hits. Opus 4.6 runs 1.67x Sonnet per token, not the 5x ratio older articles quote from pre-4.6 pricing.

If you route through a gateway for cost tracking or fallbacks, Claude Code points at LiteLLM with two env vars: ANTHROPIC_BASE_URL=https://litellm-server:4000 and ANTHROPIC_AUTH_TOKEN=sk-litellm-static-key. Setup details in Claude Code + LiteLLM. Note that claude.ai-login auth in third-party SDK products is not permitted; gateways and the Agent SDK use API keys.

Make Your Limits Last Longer

Cut the fixed overhead

• Keep CLAUDE.md under 200 lines (Anthropic's own guidance) and audit @-imports; each one loads at launch and consumes context. See the CLAUDE.md guide.
• Scope rules with paths frontmatter. A rule in .claude/rules/ with paths: ["src/api/**/*.ts"] loads only when Claude reads matching files instead of every session.
• Remove MCP servers you are not actively using. Their tool definitions ride on every turn.
• Use /compact when history grows, and start fresh sessions for unrelated tasks. More in context compaction.

Match the model to the task

• Haiku 4.5 is one third of Sonnet's price per token; Opus 4.6 is 1.67x Sonnet. Reserve Opus for architecture and multi-file coordination.
• Skip extended thinking for mechanical edits. Thinking tokens count as output.

Offload search and edits to cheaper models

File-by-file search is the dominant hidden cost: the agent reads 10-20 files into context to find one function, and every byte counts against your session. WarpGrep runs codebase search on a dedicated trained model that finds target code in 3.8 steps on average (0.73 F1), so the search tokens never touch your Claude quota. Pricing: $0 for 100k requests free, $1 per 1M requests on Pro.

The same logic applies to applying edits. Morph Fast Apply merges edit snippets at 10,500 tok/s on morph-v3-fast, so Claude emits a short lazy diff instead of rewriting whole files as output tokens. Less output per edit, more edits per session.

Frequently Asked Questions

What are Claude Code's usage limits in 2026?

A rolling five-hour session limit plus separate weekly limits, shared between Claude Code and Claude chat. The five-hour limits were doubled on May 6, 2026 for Pro, Max, Team, and seat-based Enterprise plans, and peak-hour reductions were removed. Anthropic does not publish fixed prompt counts; capacity scales by plan multiplier.

What are Claude Pro usage limits?

Pro ($17/mo billed annually, $20/mo monthly) includes Claude Code with the baseline session allowance, doubled as of May 6, 2026. Usage is shared with Claude chat. The free plan does not include Claude Code at all.

How much usage do Max 5x and Max 20x get?

Max 5x ($100/mo) gets 5 times Pro's per-session usage; Max 20x ($200/mo) gets 20 times. Both carry two weekly limits: one across all models and one for Sonnet models only.

Did Anthropic raise Claude Code limits recently?

Yes, on May 6, 2026: five-hour rate limits doubled, peak-hour reductions removed, Opus API rate limits raised. The compute came from a SpaceX deal for 300+ MW (220,000+ NVIDIA GPUs) at Colossus 1. Articles describing limit cuts predate this change.

Do Claude Code and Claude.ai share limits?

Yes. Pro and Max usage is one pool across Claude chat and Claude Code. Heavy chat use during the day leaves fewer Claude Code tokens in the same session.

How do I check how much usage I have left?

/usage or /status inside Claude Code, or Settings > Usage on Claude.ai for the combined dashboard with per-limit weekly reset times.

Does claude -p count against my session limits?

From June 15, 2026, no. Non-interactive usage (Agent SDK, claude -p, GitHub Actions, third-party apps on your subscription) draws from a separate monthly credit: $20 on Pro, $100 on Max 5x, $200 on Max 20x. When it runs out, usage flows to usage credits at API rates if enabled, otherwise requests stop. Credit does not roll over.

What should I do when I hit the limit mid-task?

Wait for the session reset (your five-hour window's end), or move to API billing: Sonnet 4.6 at $3/$15 per MTok with no session caps. Longer term, cut fixed overhead (CLAUDE.md size, MCP servers, auto memory) and offload search to WarpGrep so your quota goes to generation, not file reading.

Related Articles

• Claude Code Pricing
• Claude Code Subscription Plans
• Claude Code Hooks
• Claude Code settings.json
• Claude Code + LiteLLM
• Context Compaction