Claude Ultra drops the 5-hour limit and keeps the weekly cliff
Anthropic removed the 5-hour session limit for verified Ultra accounts on 30 October 2026 but kept an unpublished weekly limit; one subscriber's October usage shows how quickly a week can be spent without the session brake.
On 30 October 2026, Anthropic renamed its $200 Max 20x plan to Ultra and removed the 5-hour session limit for verified Ultra accounts. The weekly limit stayed. With no session brake, a week’s allowance can now be spent as fast as the work runs, and one Ultra-tier subscriber’s October logs show that a busy week’s usage fits in about 20 hours at peak pace.
Checked 31 October 2026. Anthropic sources retrieved 18:00 to 18:12 UTC; local usage measured at about 18:10 UTC. Anthropic’s statements are [vendor], measurements from this subscriber’s machine [reproduced], and analysis [inferred]. Comments from Hacker News and press quotes are attributed by name and treated as reports, not facts.
What Anthropic changed on 30 October
The announcement came on X at 17:29 UTC: “We’re renaming our Max plans: Max 5x → Max, Max 20x → Ultra. Max is still $100/month, and Ultra is still $200… We’re also removing the 5-hour limit for verified Ultra accounts”. Claude Code’s creator, Boris Cherny, repeated the limit change seven minutes later. No post appeared on Anthropic’s news page or in the Help Center release notes (opens in a new tab); the help articles themselves were rewritten about 20 minutes before the announcement. [vendor] (@claudeai (opens in a new tab), @bcherny (opens in a new tab))
The announcement as rendered by X’s official embed, captured 31 October 2026 at 18:32 UTC. The embed truncates the post at “Show more”.
The pricing page (opens in a new tab) now reads:
| Limit | Free | Pro | Max ($100) | Ultra ($200) |
|---|---|---|---|---|
| Every 5 hours | Limited | 1x | 5x | No 5-hour limit, for verified accounts |
| Every week | Limited | 1x | 5x | 10x |
The Ultra help page (opens in a new tab) describes the benefit as “Work at your own pace within your weekly limit”, and says Ultra “includes 10 times the Pro plan’s usage every week”. Accounts are verified automatically “using factors like your account and payment history”; until then “a temporary limit applies”. The same limits cover claude.ai, Claude Code and Claude Desktop. [vendor] (usage limits (opens in a new tab))
The ceiling stayed, and nobody can see its height
Anthropic publishes no number for the weekly limit: no hours, tokens or messages. Subscribers see percentage bars in Settings > Usage, and “a notification when you’re approaching your session limit or one of your weekly limits”. [vendor] (best practices (opens in a new tab), usage credits (opens in a new tab))
That makes the new “10x” impossible to compare with the old plan. The Max help page, as archived on 23 October (opens in a new tab), defined Max 20x as “20 times the Pro plan’s per-session usage allowance” and gave no weekly multiplier at all. Anthropic’s own pages therefore cannot show whether an Ultra week is larger, equal to or smaller than a Max 20x week. [vendor]
On Hacker News (opens in a new tab), the reading split. One user called it “a masterful sleight of hand” because “Max 20x is now becoming only 10x”; another replied that “the 20x plan always had only twice the weekly usage of the 5x plan. If anything this is more honest.” Neither claim can be checked against a published number, which is the problem. [inferred]
Both comments as shown on Hacker News item 50308320, captured 31 October 2026 at 18:32 UTC.
How fast a week can go: one subscriber’s October
The subscriber behind this site’s ccusage investigation pays $200 a month for Claude, the plan now called Ultra. [reported] ccusage 20.0.41 groups Claude Code’s local logs into 5-hour blocks, the same windows the old session limit used:
bunx --no-install ccusage blocks --since 20261001 --until 20261030 --timezone UTC --jsonThe export covers 59 active blocks between 7 and 30 October. Only aggregates are shown; no prompts or project names. [reproduced]
| Week (UTC) | Active blocks | Tokens | Heaviest block’s share | Two heaviest blocks’ share |
|---|---|---|---|---|
| 7–11 Oct | 19 | 4,432,917,322 | 17.5% | 30.8% |
| 12–18 Oct | 15 | 13,305,584,788 | 23.7% | 47.1% |
| 19–25 Oct | 13 | 3,572,645,298 | 18.3% | 35.7% |
| 26–30 Oct | 12 | 4,881,209,697 | 22.6% | 42.6% |
In the busiest week, ten hours of work carried 47.1% of the week’s tokens. The heaviest block, 03:00 to 08:00 UTC on 14 October, processed 3.3 billion tokens, about 661 million an hour. At that pace, the whole week’s 13.3 billion tokens would take 20.1 hours. [reproduced]
These are tokens, not Anthropic’s meter. Most are cache reads, which Anthropic may weight differently, and the size of the weekly limit in tokens is unknown. The numbers show how bursty real agent work is; they do not show where the limit would stop it. [inferred] A Hacker News user reported hitting the weekly limit “in 2 days for the first time”, and Notebookcheck (opens in a new tab) warned that a subscriber who burns “through the weekly budget in two long days” waits for the reset. Both are reports, not measurements.
The comment as shown on Hacker News item 50308320, captured 31 October 2026 at 18:32 UTC.
The brake was also a pacing tool
The 5-hour window reset “in full five hours” after the first message. It capped how much of the week any single sitting could take. Removing it does not add usage; it removes the only limit that spread consumption across the week. [vendor] (usage limits (opens in a new tab)); [inferred]
The weekly reset makes the trade sharper. It does not roll over seven days: it happens “at a fixed time each week that is assigned to your account”, whatever day the subscriber started. A heavy Monday on an account that resets on Sunday leaves six days of waiting, and the subscriber does not choose the reset day. [vendor]; [inferred]
Anthropic’s Fable models add a second meter. On Max and Ultra it can use “up to 50% of your weekly usage limits”, and its models “use them faster than other Claude models”. An Ultra subscriber who starts the week on Fable can spend half the week on one model before switching. [vendor] (Fable on your plan (opens in a new tab))
The pacing now sits with the subscriber, using a percentage bar with no unit. Agents running unattended do not read that bar. [inferred]
After the cliff, the meter switches to API prices
When any limit runs out, usage credits (opens in a new tab) let the subscriber “switch to consumption-based pricing at standard API rates”. Usage “continues until you reach your monthly spend limit”, and that limit can be “Set to unlimited”. Usage credits must be enabled and prepaid. [vendor]
The plan’s other perk does not help here. Ultra includes $200 a month of API credits (opens in a new tab), introduced on 7 October, not with this change, but they do not apply to “Interactive Claude Code in the terminal, IDE, desktop, or web” or to “Extra usage in Claude, Claude Code, or Claude Cowork.” A subscriber who empties the week on Tuesday pays API rates for the rest of it. [vendor]; [inferred]
For a heavy week, that difference is large. ccusage estimates this subscriber’s October Claude Code activity at $7,322.04 at API prices, beside a $200 plan. The estimate is not a bill, but it shows the scale of the API rates waiting on the other side of the limit. [reproduced]
What the change does fix
Long jobs no longer stop mid-task because a 5-hour window closed, which suits the pricing page’s own pitch for Ultra, “For your longest tasks”. Subscribers who work in bursts rather than evenly gain flexibility. Writing the weekly multiplier on the pricing page is also more information than the old per-session wording gave. [vendor]; [inferred]
Anthropic says the verification step exists “To help prevent fraud and abuse”. It gives no reason for removing the session limit, and it published no figure for the weekly limit, the one number that would let subscribers plan around it. [vendor]
Verdict
The change trades a frequent, visible limit for an infrequent, invisible one. For steady users, little changes. For bursty users, the week can now end in a day or two, the reset day is assigned rather than chosen, and the way past it is billed at API rates. Until Anthropic publishes what “10x Pro every week” means in a unit subscribers can measure, “work at your own pace” leaves them with no way to tell how fast that pace can be. [inferred]
Sources
- Anthropic on X, 30 October 2026: @claudeai announcement (opens in a new tab), 17:29 UTC; Boris Cherny (opens in a new tab), 17:36 UTC. The @claudeai post was read through X’s public embed, which truncates its text.
- Claude pricing, retrieved 31 October 2026: claude.com/pricing (opens in a new tab); archived 28 October (opens in a new tab), before the change.
- Claude Help Center, articles modified 30 October 2026: What is the Ultra plan? (opens in a new tab); What is the Max plan? (opens in a new tab) and its 23 October archive (opens in a new tab); How do usage and length limits work (opens in a new tab); Usage limit best practices (opens in a new tab); Manage usage credits (opens in a new tab); Monthly API credits (opens in a new tab); Fable models on your plan (opens in a new tab); Release notes (opens in a new tab), newest entry 21 October.
- Local measurement: ccusage 20.0.41,
blocksreport for 1–30 October 2026 UTC, Claude Code logs of one $200-plan subscriber, run 31 October 2026. Estimated costs are ccusage’s API-price calculation, not billing. - Captures: the @claudeai post through X’s official embed and three Hacker News comments from the public thread, taken 31 October 2026 at 18:32 UTC; originals and SHA-256 checksums are kept with the research notes.
- Reports: Hacker News discussion 50308320 (opens in a new tab), 30–31 October 2026; Notebookcheck (opens in a new tab), 31 October 2026.
- Lead only: explainx.ai (opens in a new tab), 30 October 2026. Its framing of API credits as part of this change is not supported: they were introduced on 7 October.
