The Five-Hour Wall Comes Down, One Account at a Time

On October 30, Anthropic emailed Max 20x subscribers: "Your Max 20x plan is now Ultra." Same price, same features, same billing date, same weekly limit. Then the line that matters to anyone who runs agents for hours: "We're removing the 5-hour limit."

Later that day, Claude Code still showed my five-hour window at 40%. That isn't a bug. The email is the announcement, not the confirmation. The limit comes off "in stages", once an account is verified, and Anthropic sends a second email when it's done. Until then, the percentage is real.

What changed, and what didn't

The pricing page footnote on Ultra reads "No 5-hour limit for verified accounts." The rest lives in the help center and in Anthropic's post on X. As of the evening of October 30, there was no post on anthropic.com/news or the Claude blog.

Ultra is not unlimited. The Ultra help article keeps the weekly limit across all models and the separate Fable limit, plus the usual fine print: "we may limit your usage in other ways, such as weekly and monthly caps or model and feature usage, at our discretion." On Ultra, a limit reset, if you have one, brings "your weekly usage limits back to full", per the help article.

The 20x never described the week

The old name described the limit that is now disappearing. Mike Overell, who leads product for Claude subscriptions, laid out the history: "When we launched our top plan in April 2025, it had a single limit: 5 hours. It gave you 20x Pro's limit, so we called it Max 20x. When we added weekly limits to Max 20x, we set them to 10x Pro's (the 5-hour limit stayed at 20x)."

The email says the same: "We're renaming the plan now because the 20x referred to the 5-hour limit, which we're removing." Overell added: "I get that some new users assumed 20x applied to both limits. We could have done better here, and I'm sorry."

Expect the headline number to be read as a cut. On Hacker News, one take was "this is just a barely obfuscated downgrade." The reply under it had it right: "it has been 10x of Pro weekly for a while with 20x being applied to the 5h limit." Max has no such gap: its 5x applies to both limits. Multiples are how these plans are sold, at OpenAI too: ChatGPT Pro 500 includes 25 times the Plus allowance. A multiple means little until you know which window it applies to.

My read: the rename is cosmetic but honest, and the removal is the substance. "20x" was true of the limit you hit in a burst and false of the limit you hit over a week. For people who run Claude Code all day, the week is the one that binds. The last wall my harness recorded, on October 27, was the weekly one.

Why the five-hour clock is going

The email gives the reason in one sentence: "More of Claude's work now runs for hours in the background, and a 5-hour limit shouldn't stop it partway." Overell's version: "so you can keep building without getting blocked mid-session." The pricing page now sells Ultra "For your longest tasks".

The five-hour window fits chat. A person types, reads and thinks between messages, so the person paces the work. An agent doesn't pause. A /goal or /loop run or several sessions in parallel can fill a window long before five hours pass, and then the work stops partway.

The two limits have been shifting for over a year:

• April 9, 2025. Max launches at $100 or $200 a month for "up to 5x or 20x more" usage than Pro (Engadget). The top plan has a single limit: five hours.
• July 28, 2025. Weekly limits announced for Pro and Max, in effect from August 28. TechCrunch names the targets: Claude Code running "continuously in the background, 24/7", and "sharing accounts and reselling access". The five-hour limits stay.
• March 26, 2026. Five-hour limits drain faster during weekday peak hours (announced on X).
• May 6, 2026. Claude Code's five-hour limits are doubled for Pro, Max, Team and seat-based Enterprise, and the peak-hour reduction is dropped for Claude Code on Pro and Max.
• September 14, 2026. Standard weekly limits in Claude Code go up 25%, permanently (@ClaudeDevs).
• October 27, 2026. Claude Code 2.1.312 starts calling the "session limit" the "5-hour limit" in its messages.
• October 30, 2026. Max 20x becomes Ultra, and the five-hour limit starts coming off verified Ultra accounts.

In 2025, round-the-clock background use was what the weekly limit was built to contain. In 2026, background work that runs for hours is the reason the five-hour limit goes. On Ultra, the five-hour window's only remaining role is guarding accounts that aren't verified yet.

Verification: there is nothing to click

The most important sentence in the email is easy to skim past: "We remove the limit once your account is verified." The rename happened on October 30 for everyone. The five-hour limit comes off one account at a time, after a check you can't see.

From the Ultra help article:

• It's automatic. "To help prevent fraud and abuse, we automatically verify Ultra accounts using factors like your account and payment history."
• Until then, a limit stays. "If your account isn't verified yet, a temporary limit applies. We remove it automatically once your account is verified, so there's nothing you need to do."
• Your phone number won't help. "Verifying your phone number at checkout is a separate step and doesn't remove this limit."
• One check tells you. "To check, go to Settings > Usage. If you don't see a Current session bar, your account is verified."

The order matters, too. Anthropic's October 30 post on X says the removal starts "today with most existing subscribers", and "Newer accounts keep a temporary limit until they're verified." Neither the post nor the help article gives a date for the end of the rollout or the size of the temporary limit.

Before my account was verified, the web page said "Ultra" next to "Your usage", and the Current session bar was still there. Claude Code's /usage said the same (reset times trimmed):

Current session: 40% used
Current week (all models): 22% used
Current week (Fable): 0% used

The account was verified later on October 30. The Current session bar left the web page, and /usage dropped its line too (trimmed the same way):

Current week (all models): 30% used
Current week (Fable): 0% used

Mind the names. One window has three: "Current session" in Settings and /usage, "a session limit that resets every five hours" in the limits article, and "5-hour limit" in Claude Code's limit messages.

Max or Ultra, now

My arithmetic, at monthly prices: the weekly allowance is exactly proportional to price. $20, $100 and $200 a month buy 1x, 5x and 10x Pro's weekly allowance. The old top tier's premium sat on the five-hour side, where twice Max's price bought four times Max's five-hour allowance. Ultra trades that for no five-hour limit at all. Ultra also lists $200 a month in API credits for the Claude Platform, against $100 on Max; per the help center, they don't apply to "Interactive Claude Code in the terminal, IDE, desktop, or web".

By the wall you actually hit:

• The five-hour wall on Ultra: gone once your account is verified.
• The five-hour wall on Max, with weekly room left: Ultra removes the wall once the account is verified, but it also sells you weekly headroom you may not need. Usage credits are the other way past it, at API rates (Overtime: When the Plan Runs Out, Credits Take Over).
• The weekly wall on Max: Ultra doubles the week at double the price.
• The weekly wall on Ultra: no relief. The week is the same size, and without the five-hour window you may reach it sooner.

Model choice still matters inside the week: per the pricing page, Fable can use "up to half your weekly limit" (Fable 5.1 Is In Your Max Plan. It Is Not Free.), and higher reasoning effort uses more of every limit (The Thinking Dial).

The week is the only clock

The good part first. On a verified Ultra account, a long run no longer stops because a five-hour window filled. In the email's words, "you can spread your weekly usage limit across the week however you like."

It is still one pool. Usage on claude.ai, Claude Code and Claude Desktop "counts towards the same usage limit", and subscription-authenticated agents share it too.

The less pleasant part: the five-hour window was a brake you didn't have to build. A loop that went wrong, or an agent stuck retrying, could burn at most one window before the five-hour wall stopped it, unless usage credits paid for more. On a verified Ultra account the first wall is the weekly limit, and a spent week stays spent until your reset.

With usage credits on, the wall becomes a bill. Anthropic's usage credits article: "Instead of being blocked when you reach one of your plan's limits, you can switch to consumption-based pricing at standard API rates and continue your work without interruption." That is also how a runaway loop turns into an invoice.

Anthropic's usage page helps with the pacing. When I wrote this, mine read: "On track. You should reach Wednesday's reset with room to spare." Later the same day: "Heads up. At this pace you'll run out Tuesday night, before Wednesday's reset." The verdict follows your pace and can flip within hours, so check it before an overnight run.

Brakes worth keeping:

• Watch the week. rate_limits.seven_day.used_percentage is the number that matters now. My harness watches both windows and wraps up cleanly before the wall (Handling AI Usage Limits and Clean Handovers). On a verified Ultra account it watches the weekly one alone, and when that wall hits, the wait can be days, not hours.
• Give every loop an exit. The Loop on a Leash covers how /loop stops. Unattended runs need a check that can say "done" and a budget that can say "enough".
• Decide about credits before the wall. Should hitting the week stop work or bill it? Settle that, and set a budget (The Cost Control Plane).

What the status line can see

Claude Code runs your status line command and passes it JSON on stdin. The docs list rate_limits.five_hour.used_percentage and rate_limits.seven_day.used_percentage, each "from 0 to 100", each with a resets_at in Unix epoch seconds. Three documented details matter:

• rate_limits appears only for claude.ai Pro, Max and Ultra subscribers, or behind a Claude apps gateway that sets a spend limit, and only after the first API response in a session.
• "Each window (five_hour, seven_day, spend_limit) may be independently absent".
• Claude Code "drops a window once its resets_at time passes".

The docs say nothing about verified Ultra accounts. Once my account was verified, the status line received seven_day with no five_hour. The documented // empty idiom handles a missing window:

#!/usr/bin/env bash
# Status line script: Claude Code passes its JSON on stdin.
input=$(cat)
five=$(jq -r '.rate_limits.five_hour.used_percentage // empty' <<<"$input")
week=$(jq -r '.rate_limits.seven_day.used_percentage // empty' <<<"$input")
line=""
[ -n "$five" ] && line="5h $(printf '%.0f' "$five")%"
[ -n "$week" ] && line="${line:+$line  }7d $(printf '%.0f' "$week")%"
printf '%s\n' "$line"

It prints 5h only when the window is present and never reads the plan name, so Pro, Max and unverified Ultra accounts keep their figure. Because the script keeps no state, that is all it needs. The trouble starts when you cache.

Follow the data, not the plan name

My Claude Code status line shows each plan window as a five-cell bar, a percentage and the local reset time. The full reference is My Self-Tuning AI Harness; The Harness: Two AIs, Zero Trust shows where it started.

A session with no response yet has no rate_limits, and neither does the account picker at launch. So the harness keeps a small per-account usage cache. Status line readings and Claude Code's own /usage report (a local command, no model tokens) fill it, with Anthropic's usage endpoint as a fallback.

That cache had a flaw the Ultra change would have exposed. A cached window past its reset time reads 0%, which is right for a plan with a five-hour limit: a new window starts empty. But nothing ever deleted a window. Once Anthropic stopped reporting the five-hour window for an account, my bar would have shown 5h 0% forever, and the account picker 5h —, which in my harness means "unknown".

The tempting fix is one line: if the plan is Ultra, hide the 5h area. It is wrong two ways:

• The removal is staged. An Ultra account that isn't verified yet still has the limit; hiding it hides a wall you can still hit.
• A label is a guess. Plan names just changed under everyone. The usage data is the account's actual state.

So the rule is: hide the 5h area only when the account's own data says its plan has no five-hour window. Decide per account, from data, and let the data change its mind.

The hard part is reading absence safely. A missing five_hour means either "this plan has no five-hour limit" or "the last window ended and this session has been idle since". The harness only concludes the first from evidence that can't be the second:

• A complete report without the row. Claude Code's /usage report has a weekly row and no "Current session" row, or the usage endpoint returns no five-hour window. That endpoint already returns null for windows a plan doesn't have.
• A fresh weekly reading without the window. In a live session, a weekly reading newer than anything recorded arrives with no five-hour window. The premise is that a fresh response on a plan with a five-hour limit carries that window. An idle session repeating old numbers is never newer than what's recorded, so the gap after a reset can't trigger it.

Either one clears the cached five-hour window and marks the account. A newer five-hour reading removes the mark, so the bar comes back on its own if the data says so. The rule leans toward showing on purpose: hiding a real limit lets you walk into a wall you stopped watching, while showing a stale one costs a few characters.

Once an account is marked, the status line drops the 5h bar and the 5h alert, and the launch picker and the accounts table stop showing 5h —. Everything else that reacts to limits runs on the weekly window alone. For an account with a five-hour limit nothing changes, including 5h 0% after a reset.

Once Anthropic verifies an account and the data stops reporting a five-hour window, the 5h area goes away by itself, with nothing to watch and no setting to change.

Bottom line

The rename fixes a label that, by Anthropic's own account, confused people. The removal changes the shape of the budget: on a verified Ultra account the week is the only plan window, and the pacing that the five-hour clock used to do is now yours.