If you've lost track of whether Claude got more generous or more restrictive lately, you're not imagining things — it's done both. Since August 2025, Anthropic has changed Claude's usage limits at least three separate times: it added weekly caps on top of the existing rolling window, ran a two-week off-peak doubling in March 2026, then permanently doubled Claude Code's 5-hour limits on May 6, 2026. Two of those changes are still active. One ended on schedule. This post lays out the dated timeline so you can tell which limit you're actually running into today, and links out to Anthropic's own numbers rather than secondhand recaps.
If you're also weighing whether your subscription tier makes sense given these changes, our Claude Code pricing guide breaks down Pro versus Max 5x versus Max 20x in more depth, and our coverage of the Claude Max usage-limits lawsuit covers the class-action angle if you feel the advertised multipliers don't match what you're getting.
Update — August 14, 2026: Claude Code desktop shipped an auto-continue checkbox that resumes a limit-blocked session automatically once your window resets — a UX fix, not a limits change. Full breakdown.
Update — July 18, 2026: 30-minute Fable 5 access glitch (Jul 17 ~18:30 UTC) — false usage credits required before the July 19 promo deadline. @ClaudeDevs refunded overcharges + matching credits. Incident recap.
The Short Version
There is no single product called "Claude 2x." Every time you've seen that phrase, it's referred to a specific, dated change to the limits on an existing plan — never a new SKU you can buy. Three events matter here:
| Date | Change | Status as of July 2026 |
|---|---|---|
| Aug 28, 2025 | Weekly rate limits added on top of the rolling 5-hour window | Still in effect |
| Mar 13–28, 2026 | Off-peak promo doubled 5-hour limits outside peak hours and on weekends | Ended, no longer active |
| May 6, 2026 | Claude Code's 5-hour limits permanently doubled; peak-hour throttling removed for Pro/Max | Still in effect |
The weekly cap is the one constant across all three events. Neither the March promo nor the May 6 change touched it.
Timeline: Every Change Since August 2025
August 28, 2025 — Weekly Caps Arrive
Anthropic announced new weekly rate limits for Claude Pro and Max on July 28, 2025, confirming on X that the change would go into effect in late August and would "apply to less than 5% of subscribers based on current usage." The limits took effect on August 28, 2025, stacking a 7-day reset cycle on top of the existing rolling 5-hour window.
Per Anthropic's published allowances at the time, weekly usage broke down roughly as follows:
| Plan | Sonnet 4 per week | Opus 4 per week |
|---|---|---|
| Pro ($20/mo) | 40–80 hours | Not included |
| Max ($100/mo) | 140–280 hours | 15–35 hours |
| Max ($200/mo) | 240–480 hours | 24–40 hours |
Anthropic's stated reasons, as reported by TechCrunch, were: a small number of users running Claude Code essentially 24/7, account sharing, and unauthorized reselling of access — all degrading availability for everyone else amid what Anthropic called "unprecedented demand." This is the moment Claude went from one limit to two, and it's the reason every later "doubling" headline has to specify which limit it touched.
March 13–28, 2026 — The Off-Peak Promo
Anthropic ran a temporary promotion doubling 5-hour usage limits during off-peak hours and all weekend for Free, Pro, Max, and Team users. Peak hours were defined as 8 AM–2 PM ET (5–11 AM PT, 12–6 PM GMT) on weekdays — outside that window, and all day Saturday and Sunday, users got double their normal 5-hour allowance. The bonus usage did not count against weekly caps. Enterprise plans were excluded.
The promo ended on schedule on March 28, 2026. If you're still budgeting your Claude Code sessions around a "2x on weekends" assumption, that window closed months ago.
May 6, 2026 — The Permanent Doubling
On May 6, 2026, Anthropic announced it had permanently doubled Claude Code's 5-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans, and removed the peak-hour limit reduction that previously applied to Pro and Max accounts. Claude Opus API rate limits also went up substantially.
The change was enabled by a new compute agreement: Anthropic secured access to SpaceX's Colossus 1 data center, described in the announcement as delivering "more than 300 megawatts of new capacity (over 220,000 NVIDIA GPUs) within the month." Anthropic also referenced parallel capacity deals with Amazon (up to 5 GW), Google (5 GW with Broadcom, starting 2027), and a $30 billion Microsoft/NVIDIA Azure agreement — context for why this doubling arrived when it did rather than earlier.
Three things this change did not touch:
- Weekly caps — unchanged, same reset cadence.
- The Free plan — excluded from the permanent doubling (Free doesn't get Claude Code access at all).
- API pricing — Anthropic's per-token rates on the API were untouched.
For a head-to-head look at how this landed against competing tools, see our coverage of the Claude Code vs. Codex rate-limit boost.
Why Anthropic Keeps Adjusting the Limits
Each of these three events maps to the same underlying variable: the gap between demand for Claude Code and the compute Anthropic has online at any given moment.
- Demand outpaces supply → limits tighten. The August 2025 weekly caps are the clearest example — a popular tool grew faster than infrastructure, so Anthropic added a structural limit rather than letting availability degrade for everyone.
- New capacity comes online → limits loosen. The May 6, 2026 doubling rode directly on the Colossus 1 deal. When headroom appears, Anthropic has generally extended it to paying tiers first.
- Abuse patterns get addressed with structure, not just message counts. Account sharing and 24/7 automated use pushed Anthropic toward a weekly cap rather than a simple per-message ceiling, because a rolling 5-hour window alone couldn't distinguish a legitimate long session from sustained background automation.
The practical implication: don't treat any single announcement as final. Anthropic has revised these numbers multiple times in under a year, and the Claude Max usage-limits lawsuit filed in June 2026 is itself evidence that even Anthropic's own marketing multipliers ("5x," "20x") have been contested in practice. Check /usage when a workflow starts hitting walls rather than relying on memorized figures.
Before vs. After the May 6 Doubling
| Plan | Pre-May 6 5-hr limit | Post-May 6 5-hr limit | Peak-hour throttle | Weekly cap |
|---|---|---|---|---|
| Pro | Baseline | 2x baseline | Removed | Unchanged |
| Max 5x | Baseline | 2x baseline | Removed | Unchanged |
| Max 20x | Baseline | 2x baseline | Removed | Unchanged |
| Free | No Claude Code access | No Claude Code access | N/A | N/A |
Anthropic doesn't publish exact message counts — it publishes plan multipliers (5x, 20x relative to Pro) and lets weekly-hour figures serve as the closest concrete anchor. If you were on Pro before May 6 and noticed sessions throttling mid-afternoon, that specific pain point — peak-hour reduction — is gone. The weekly bucket, though, is the same size it's always been; if a sustained multi-day workload was your actual bottleneck, this change doesn't move that number.
The Structure That Hasn't Changed: 5-Hour Window vs. Weekly Cap
Underneath every headline, Claude enforces two independent limits, and understanding both is what actually keeps you from getting surprised:
- A rolling 5-hour window that resets continuously and protects against short, intense bursts.
- A 7-day weekly cap, introduced in August 2025, that protects against sustained heavy use across a full week.
On Max plans, Sonnet and Opus draw from separate 5-hour and weekly buckets. On Pro, the two models share a single pool — so running a long Opus session on Pro eats into your Sonnet headroom for the rest of the window, while on Max it doesn't.
Check both numbers directly rather than guessing:
$ claude /usage
The output shows 5-hour consumption per model, weekly cap percentage, and reset times. This is the only figure that reflects Anthropic's current configuration — community trackers and single-point-in-time articles (including this one) are a snapshot of one date, not a live feed.
If you're regularly running into the context side of these limits rather than the usage side, our guide on managing Claude Code's context window covers the /compact and /clear workflow that keeps sessions efficient regardless of which plan you're on.
Pro vs. Max 5x vs. Max 20x: Picking a Plan Post-Doubling
| Plan | Price | Best for | Notes after May 6 |
|---|---|---|---|
| Pro | $20/mo | Assistant-style chat, light Claude Code use | No more peak-hour throttling; doubled 5-hr window |
| Max 5x | $100/mo | Daily Claude Code sessions | 5x Pro's headroom, both models on separate buckets |
| Max 20x | $200/mo | Heavy agentic workflows, multi-repo sessions | Highest request priority; still weekly-capped like every other tier |
If Pro's 5-hour window was your bottleneck before May 6, it's worth re-testing before assuming you still need Max — the removed peak-hour throttle alone changes a lot for anyone whose sessions used to run during the 8 AM–2 PM ET window. For a fuller cost breakdown including bring-your-own-API-key economics, see the Claude Code pricing guide.
Five Ways to Stretch Any Plan's Limits
Regardless of which numbers are current when you read this, these habits reduce how fast you burn through either limit:
- Route by task, not by default model. Plan with Opus, execute with Sonnet, review with Opus. Sonnet burns through allowance far slower per turn than Opus, and most execution steps don't need Opus-level reasoning.
- Use
/compactand/cleardeliberately. Every turn in a long session re-sends prior context. Compacting periodically and clearing between unrelated tasks keeps token consumption — and by extension your window burn — proportional to actual work done. - Move stable context into a project-level CLAUDE.md. Cached system context is discounted; repeatedly pasting the same project conventions into prompts is not.
- Plan before you build. Ask for a plan, approve it, then execute. Correcting a wrong plan mid-execution costs more turns than getting the plan right up front.
- Batch related questions into one turn. Three separate "what about this error?" messages cost three times what one combined message does, for the same answer quality.
Related Reading
- Claude outage — two network failures cut capacity, July 29-30, 2026
- Claude Reflect — usage patterns vs rate limits (July 9)
- Claude Code Pricing Guide 2026 — full Pro vs. Max vs. API-key cost breakdown
- Anthropic Sued Over Claude Max Usage Limits — the June 2026 class-action over advertised multipliers
- Claude Code vs. Codex: Rate Limit Boost — how the May 6 change shifted the competitive picture
- Claude Programmatic Usage Credits — the Agent SDK billing walkback from the same period
- Managing Claude Code's Context Window —
/compactand/clearworkflows that stretch any plan further - AI Subscription True Cost: ChatGPT vs. Claude vs. Gemini — how Claude's pricing compares industry-wide
- Official source: Anthropic — Higher Usage Limits for Claude and a Compute Deal with SpaceX
Usage limits and plan allowances change frequently. The figures above reflect Anthropic's published announcements as of July 5, 2026 — always confirm current numbers with /usage or claude.ai before making a plan decision.