Claude Usage Limits in 2026: Every "2x" Change, Explained
If you've lost track of whether Claude's limits went up, down, or sideways, you're reading the right page. Since August 2025, Anthropic has reworked Claude's usage limits at least three times: it added weekly caps, ran a two-week off-peak "2x" promo in March 2026, then permanently doubled Claude Code's 5-hour limits on May 6, 2026. Some of those changes ended. Some are still live. This post is the running map, last checked June 26, 2026, so you can tell which limit you're actually hitting today.
If Claude is blocking access before you reach a usage cap, check the separate Claude identity-verification rules; that account gate is not a rate limit.
Key takeaways:
- Claude's limits aren't static. Anthropic has changed them repeatedly since the August 28, 2025 weekly-cap rollout and keeps tuning them as compute supply shifts.
- The March 13–28, 2026 promo was a temporary off-peak boost. It ended and is no longer active.
- On May 6, 2026, Anthropic permanently doubled Claude Code's 5-hour limits for Pro, Max, Team, and seat-based Enterprise plans.
- Across every one of these changes, the weekly cap and the two-limit structure stayed the same. Master the structure once and each new announcement is easy to place.
What "Claude 2x Usage" Actually Means (And Why It Keeps Coming Back)
"Claude 2x usage" is shorthand people use for any moment Anthropic doubles a usage allowance, and it has referred to different things at different times. In 2026 alone it meant two separate events: the March 13–28 temporary off-peak promotion that doubled 5-hour limits during off-peak hours, and the May 6 permanent doubling of Claude Code's 5-hour rate limits for paying plans, driven by Anthropic's SpaceX/xAI Colossus 1 compute deal.
Here's why the phrase stays confusing: there has never been a permanent product called "Claude 2x." Every "2x" has been a change to the limits on existing plans (Free, Pro, Max, Team, Enterprise), not a new SKU you can buy. So when someone says "Claude doubled its limits," the honest first question is which doubling, and when.
The cleanest way to keep the events apart is a road analogy. The March promo was a temporary HOV lane, open off-peak and weekends, gone after March 28. The May 6 change is a permanent extra lane on the freeway, always open, but only on paying plans. And the weekly cap? That's the speed limit. It didn't move in either case.
So when you hear Claude's limits "doubled," anchor to the date. March means a promo that ended. May 6 means a permanent Claude Code change that's still live as of this update.
Claude's Usage-Limit Timeline: Every Major Change Since 2025
The fastest way to stop being confused by each new headline is to see the whole sequence at once. Anthropic hasn't made one big limit decision, it's made a series of them, each responding to the same underlying tension between demand and available compute.
| Date | What changed | Type | Still in effect? |
|---|---|---|---|
| Aug 28, 2025 | Weekly caps added on top of the existing 5-hour window, aimed at non-stop Claude Code use and account-sharing | Restriction | Yes, weekly caps remain |
| Mar 13–28, 2026 | Off-peak promo doubled 5-hour limits off-peak and all weekend, for Free/Pro/Max/Team | Temporary boost | No, ended Mar 28 |
| May 6, 2026 | Claude Code 5-hour limits permanently doubled for Pro/Max/Team/seat-based Enterprise; peak throttling removed for Pro/Max | Permanent boost | Yes |
| Through mid-2026 | Continued capacity-driven tuning as the Colossus 1 deal came online | Ongoing adjustment | Varies, check /usage |
A few things jump out of that table. First, the weekly cap is the through-line, introduced in 2025 and untouched by every "2x" since. Second, the changes alternate between tightening (2025) and loosening (2026) as Anthropic's supply situation flips. Third, the only reliable source of truth for your current numbers is the /usage command, not any single news story, because the next adjustment may have already landed by the time you read one.
The August 2025 Weekly Caps: Where This Story Starts
Most "2x" coverage skips the origin. In late July 2025, Anthropic announced new weekly rate limits that took effect August 28, 2025, sitting on top of the existing rolling 5-hour window. The stated reason was a small group of users running Claude Code essentially 24/7, plus account-sharing and reselling, which degraded availability for everyone else. That's the moment Claude went from one limit to two, and it's why every later "doubling" has to specify which limit it touched.
March 2026: The Off-Peak Promo
The March 2026 promotion ran March 13 through March 28, doubling 5-hour usage limits during off-peak hours and all day on weekends for Free, Pro, Max, and Team users. Peak hours were 8 AM – 2 PM ET (5–11 AM PT, 12–6 PM GMT) on weekdays, per Anthropic's Help Center. Crucially, the bonus did not count toward weekly limits. Enterprise was excluded; everyone else got the boost.
May 6, 2026: The Permanent Doubling
On May 6, 2026, Anthropic permanently doubled Claude Code's 5-hour rate limits for Pro, Max, Team, and seat-based Enterprise customers, and removed peak-hour throttling for Pro and Max. This one's covered in depth below, because it's the change most people are still feeling.
Why Anthropic Keeps Changing Claude's Limits
If it feels like the limits move every couple of months, that's because the thing underneath them moves every couple of months: the gap between how much people want to use Claude and how much compute Anthropic has online. Understanding that one dynamic explains every past change and lets you predict the shape of the next one.
Three forces drive the adjustments:
- Demand outruns supply, so limits tighten. When a popular model or a tool like Claude Code takes off faster than data-center capacity, Anthropic protects availability by adding or lowering caps. The August 2025 weekly limits are the textbook example.
- New capacity comes online, so limits loosen. When a compute deal closes, headroom appears and Anthropic hands it to paying tiers first. The May 6 doubling rode directly on the Colossus 1 capacity.
- Abuse gets curbed, so rules get specific. Account-sharing, reselling, and 24/7 automation push Anthropic toward structural limits (like the weekly cap) rather than simple message counts.
The practical takeaway: don't treat any single announcement as the final word. Treat Claude's limits as a moving number tied to compute economics, learn the structure once, and re-check /usage whenever a workflow starts hitting walls. That habit ages far better than memorizing this month's figures.
The May 6, 2026 Permanent Doubling (And the SpaceX Deal Behind It)
On May 6, 2026, Anthropic permanently doubled Claude Code's 5-hour rate limits for Pro, Max, Team, and seat-based Enterprise customers, and removed peak-hour throttling for Pro and Max. The change was enabled by a new SpaceX/xAI Colossus 1 compute deal: 300+ MW and 220,000+ NVIDIA GPUs coming online within the month. Weekly caps were unchanged.
What Actually Changed on May 6
Three concrete things shifted:
- Claude Code's 5-hour limits doubled for Pro, Max, Team, and seat-based Enterprise.
- Peak-hour throttling was removed for Pro and Max, no more reduced limits between 8 AM and 2 PM ET.
- The Free plan was excluded. Free users got the March promo but do not get the permanent doubling.
The Free plan got the March promo but was deliberately excluded from the May 6 doubling, Anthropic's compute is going to paying tiers first.
Why Then? The SpaceX/Colossus 1 Compute Deal
Anthropic had been throttling for months because demand outran supply. The Colossus 1 deal, 300 megawatts of capacity and over 220,000 NVIDIA GPUs becoming available within the month, finally gave them the headroom to lift the cap on Claude Code without hitting infrastructure walls. More compute also means more breathing room for the heavier models, which matters if you're using Opus 4.7 for planning and review steps.
9to5Google's coverage framed it as the first concrete capacity win for paying Claude Code users in 2026, and that's a fair read.
What Stayed the Same
Three things did not change on May 6:
- Weekly caps, same numbers, same reset cadence.
- API pricing, Anthropic's API tiers were untouched.
- Model availability, Sonnet, Opus, and Haiku availability across plans is the same as it was on May 5.
Before vs After: Claude Code Limits Side-by-Side
Before May 6, 2026, Claude Code on Pro had reduced limits during peak hours and roughly half the post-May-6 5-hour throughput. After May 6, Pro and Max plans get full peak-hour access plus doubled 5-hour windows. Weekly caps stayed the same on every plan.
| Plan | Pre, May 6 5-hr¹ | Post, May 6 5-hr¹ | Peak throttle (pre) | Peak throttle (post) | Weekly cap |
|---|---|---|---|---|---|
| Pro | ~45 Opus / ~225 Sonnet | ~90 Opus / ~450 Sonnet | Yes (reduced) | None | Unchanged |
| Max 5x | ~225 Opus / ~1,125 Sonnet | ~450 Opus / ~2,250 Sonnet | Yes (reduced) | None | Unchanged |
| Max 20x | ~900 Opus / ~4,500 Sonnet | ~1,800 Opus / ~9,000 Sonnet | Yes (reduced) | None | Unchanged |
| Free | (no Claude Code) | (no Claude Code, excluded) | n/a | n/a | n/a |
¹ Anthropic publishes ratios (5x, 20x), not message counts. Numbers reflect community testing and approximate doubling, and may shift with later adjustments.
In plain English: if you were on Pro before May 6 and noticed your Claude Code session running out of room mid-afternoon, the limit you were hitting moved up. Roughly twice the room, no peak-hour penalty. The weekly bucket is still the same size, so if your workload sustained over a full week was the bottleneck, May 6 didn't fix that.
We're running Max 5x across our team and felt the May 6 change inside an hour, peak-hour Claude Code sessions stopped throttling mid-prompt. If you canceled Pro in February because peak hours felt useless, May 6 reset the math. It's worth re-evaluating against Claude Code's 2026 roadmap before assuming the same calculus applies.
What We Measured: A Real Higgsfield-Heavy Workload Through the New Limits
We re-ran our /refresh-images job on May 7, 2026, the day after the permanent doubling, and it became the cleanest before/after test we've had this year. The workload: regenerate hero + inline images for the top 10 techsy.io posts via the Higgsfield MCP server (Nano Banana Pro / GPT Image 2), patch 32 published Sanity docs in place, and ping IndexNow on all 30 affected URLs. One Claude Code session, Max 5x, Sonnet primary with Opus on the planning step.
The post-doubling run, measured:
| Metric | Value |
|---|---|
| Posts refreshed | 10 |
| Images generated (hero + inline) | ~30 |
| Sanity docs patched | 32 |
| Higgsfield credits spent | 243 (~$15) |
| Wall-clock | Single Claude Code session, no rate-limit interruption |
| 5-hr window consumed | ~38% of post-May-6 Max 5x |
The same workload pre-May-6 would have eaten roughly 76% of the old Max 5x window by the published ratios, uncomfortably close to the cap once you factor in peak-hour throttling between 8 AM and 2 PM ET, which compresses sustained MCP loops well below the headline message count. We tried similar batches in late April and capped out around post 5 or 6, mid-image-generation. Same script, same MCP server, half the throughput.
Why MCP-heavy sessions feel the change more than chat does
A pure conversational Claude session uses 1 turn per user message. An MCP-driven loop spends 5–10 Sonnet turns per "thing done", a Higgsfield generation alone is start_job → poll → poll → poll → fetch → upload → patch_reference. Each of those turns counts against the same 5-hour bucket, so the bucket drains 5–10x faster than chat. Doubling the bucket roughly doubles how many MCP "things" finish before the cap.
Concretely, on Max 5x post-doubling we now expect:
- ~225 Higgsfield generations per 5-hour window (was ~110 pre-May-6) at our measured ~10-turn-per-image cost
- ~75 full-post image refreshes per window (was ~37) including Sanity reference patching
- No mid-session throttle dropouts between 8 AM and 2 PM ET, the part of the day this kind of batch usually runs
These are calculated from Anthropic's published ratios (Max 5x: ~2,250 Sonnet / 5-hr post-doubling), not extrapolated. Your mileage shifts with how chatty your MCP server is, a faster-completing job means fewer poll turns, which stretches the window further.
The headline isn't that 2x is exactly 2x. It's that for tool-heavy agent workflows, MCP, sub-agents, plan-then-build cycles, the practical ceiling moved from "this batch will probably finish, but cut it close" to "this batch finishes with room to spare." That's the qualitative change behind the quantitative doubling.
Pro vs Max 5x vs Max 20x: Choosing the Right Plan
Plans and prices can change too, so treat the numbers below as the mid-2026 baseline and confirm against Anthropic's current pricing before you upgrade. The decision logic, though, is durable: match the plan to how hard you lean on Claude Code.
| Plan | Monthly price | ~Sonnet messages / 5-hr¹ | ~Opus messages / 5-hr¹ | Weekly cap | Best for |
|---|---|---|---|---|---|
| Pro | $20 | ~450 | ~90 | Standard | Assistant chat + light Claude Code |
| Max 5x | $100 | ~2,250 | ~450 | 5x Pro | Daily Claude Code sessions |
| Max 20x | $200 | ~9,000 | ~1,800 | 20x Pro | Heavy agent / multi-repo workflows |
¹ Approximate, based on community testing post-May-6. Anthropic publishes ratios only.
Pro: Who It's For Now
Pro at $20/month covers most users who treat Claude as a smart assistant, code questions, writing help, the occasional refactor. With peak-hour throttling gone and 5-hour limits doubled, Pro is a much stronger "starter for serious users" than it was in February. If you're not running Claude Code daily for hours at a stretch, this is probably enough.
Max 5x: The Sweet Spot for Daily Claude Code
If your daily Claude Code session ends at 3pm with a rate-limit message, you're a Max 5x candidate. At $100/month, you get roughly five times Pro's headroom across both Sonnet and Opus, and the math starts working out the moment Claude Code becomes a daily part of how you ship.
Max 20x: Heavy Agent / Multi-Repo Workflows
Max 20x at $200/month is for heavy agent workflows, long-running Claude Code sessions across multiple repos, multi-step plans, or anyone running Claude Code as part of a CI-adjacent loop. If you're already at Max 20x and still hitting walls, the next step is usually API-side cost optimization rather than another upgrade. And if these limits still aren't enough, comparing Claude Code with Cursor and Copilot is the right next stop.
The 5-Hour Window vs the Weekly Cap: The Structure That Never Changes
This is the part worth memorizing, because it has survived every "2x" announcement. Claude enforces two separate limits: a rolling 5-hour window that resets continuously, and a weekly cap that resets every 7 days. The 2026 doublings touched the 5-hour window only. The weekly cap, in place since August 2025, stayed put. On Max plans, Sonnet and Opus run on independent buckets.
Why two limits? The 5-hour window protects Anthropic's infrastructure from short bursts (somebody firing 800 prompts in twenty minutes). The weekly cap protects against sustained heavy use over a full week. They serve different purposes, and that's why doubling one doesn't automatically double the other.
Doubling the 5-hour window without doubling the weekly cap means you can sprint harder, but the marathon distance didn't change.
You can check both in real time with the /usage command, and this is the habit that keeps you current no matter what Anthropic changes next:
$ claude /usage
Plan: Max 5x
5-hour window (Sonnet): 312 / ~1,125 messages used (resets in 2h 14m)
5-hour window (Opus): 47 / ~225 messages used (resets in 2h 14m)
Weekly cap: 68% used (resets Sun 00:00 UTC)(Numbers approximate and illustrative. Anthropic publishes ratios, not message counts.)
On Max plans, Sonnet and Opus each have their own 5-hour and weekly buckets, per Anthropic's Nov 24, 2025 update. On Pro, they share. That distinction matters when switching between Sonnet and Opus inside a single session, on Max, the cost is "spent against whichever model you used"; on Pro, it's a shared pool.
How to Stretch Your Limits Without Upgrading
Whatever the current numbers are, these five habits get more out of any plan, and they keep working through every limit change.
1. Smart routing, Opus-Sonnet-Opus pipeline. Plan with Opus, build with Sonnet, review with Opus. Opus is great at thinking and reviewing; Sonnet is great at executing the plan. On our own daily workflow, switching from default-Opus to the Opus-Sonnet-Opus smart routing pattern roughly tripled how much we could ship before hitting the 5-hour cap.
2. /clear and /compact aggressively. Long context windows are expensive, every turn re-sends everything before it. Hit /compact every ~30 turns to summarize history into a short briefing, and /clear when you're truly starting a new task. Most users hold context far longer than they need to.
3. Cache structured CLAUDE.md context. Stable system context, the kind you'd put in a structured CLAUDE.md context, gets cached at roughly a 90% discount. Move project conventions, file maps, and style rules into CLAUDE.md instead of pasting them into prompts. The savings compound.
4. Plan-then-Build mode. Don't let Claude Code freelance. Ask it for a plan first, approve the plan, then ask it to execute. Claude is much faster at executing a plan than at "figuring it out", and you spend fewer messages re-correcting. Combining this with Claude Code hooks to auto-enter Plan mode on certain commands is a small but compounding win.
5. Batched prompts and queued questions. Group three small questions into one turn. "Walk me through these three errors and tell me which is highest priority" is one message; three separate "what about this error?" turns is three. Same answer quality, one-third the budget. Anthropic's own best-practices doc is worth a quick skim if you want the official version.
Frequently Asked Questions
Why does Anthropic keep changing Claude's usage limits?
Because the limits track a moving target: the gap between demand and available compute. When usage outruns capacity, Anthropic tightens limits (the August 2025 weekly caps); when a new compute deal closes, it loosens them (the May 2026 doubling). Expect periodic adjustments rather than one fixed number, and check /usage for your current figures.
Did Claude permanently double its limits?
Yes for Claude Code on Pro, Max, Team, and seat-based Enterprise plans as of May 6, 2026, the 5-hour rate limits were permanently doubled and peak-hour throttling was removed for Pro and Max. No for the weekly cap, no for the Free plan, and no for non-Code chat usage on claude.ai. The doubling is specifically a Claude Code change. (Anthropic Help Center)
Is the March 2x promo still active?
No. The March 13–28, 2026 off-peak promo ended on March 28 and is no longer active. The May 6 permanent doubling is the change still in effect. Two different events, only one of which is current.
When did Claude get weekly limits?
Anthropic announced new weekly rate limits in late July 2025, effective August 28, 2025, on top of the existing 5-hour window. The stated goal was curbing a small number of users running Claude Code nonstop, plus account-sharing and reselling. Every "2x" event since has left this weekly cap untouched.
Do my weekly limits also double?
No. Anthropic was explicit: weekly caps did not change in the 2026 events. Only 5-hour windows changed. The March promo bonus didn't count toward weekly, and the May 6 doubling left the weekly cap untouched. So if you're hitting the weekly limit on day 5, neither announcement helps. That's by design, the two limits exist for different reasons.
What were Claude's "peak hours"?
During the March promo, peak hours were 8 AM – 2 PM ET (5–11 AM PT, 12–6 PM GMT) on weekdays, those were the windows where the 2x bonus did not apply. Weekends were 2x all day. After May 6, peak-hour throttling was removed entirely for Pro and Max on Claude Code, so the concept of "peak hours" no longer reduces your available limits on those plans.
Did the Free plan get the May 6 doubling?
No. Free users got the March promo but were excluded from the permanent doubling. Anthropic restricted the May 6 change to paying tiers, Pro, Max, Team, and seat-based Enterprise. Free users still have access to claude.ai with standard limits, but no Claude Code access and no share of the new compute capacity from the SpaceX/Colossus 1 deal.
How do I check how much usage I have left?
In Claude Code, run /usage. It prints your current 5-hour window consumption (separately for Sonnet and Opus on Max plans), your weekly cap percentage, and reset times. On the web UI, the /usage panel shows the same data. Treat the /usage output as the source of truth, especially because limits change, community trackers like claude2x.com only reflected the March promo and are no longer authoritative.
Are Sonnet and Opus limits shared or separate?
On Max plans, Sonnet and Opus run on independent 5-hour buckets and independent weekly buckets, per Anthropic's Nov 24, 2025 update. On Pro, they share buckets. So on Max, burning your Opus quota doesn't affect Sonnet, you can keep working on Sonnet for the rest of the window. On Pro, every Opus message reduces your Sonnet headroom too.
Should I upgrade from Pro to Max 5x?
If you hit Pro's 5-hour cap regularly during a normal Claude Code session, community testing suggests this happens around 45 Opus or 225 Sonnet messages pre-May-6, roughly doubled now, Max 5x roughly 5x's that headroom for $100/month. Heavy agent users running multiple repos or long Plan-then-Build cycles should skip directly to Max 20x. Most "Pro feels tight" users land on Max 5x.
Building something on top of Claude that needs to scale beyond these limits? Get a free AI integration consultation.