Policies and sources checked through November 6, 2026. If the five-hour window regularly interrupts work while weekly capacity remains, verified Ultra may improve its rhythm. If the problem is unacceptable answers, unavailable models or an exhausted weekly pool, address that cause first.
This guide is for current Max users and people choosing between Max and Ultra. The logs, prompts and economic examples are original illustrations. No model calls or account tests were performed. They provide a recording and decision method, without guaranteeing performance or verification.
1. What the new names mean: identify capacity and the purchase in question
The former Max 5x is Max, and Max 20x is Ultra. The readable official announcement and current help pages confirm the rename. Existing 20x subscribers move automatically to Ultra without a price change. The original post was published at 17:29 UTC on October 30, 2026, early October 31 in Singapore. This is not a new model launched on November 6.
Ultra and automatic verification · Official Claude Max rename announcement
Current US web prices are $100/month for Max and $200/month for Ultra. Max offers 5× Pro usage both per five-hour window and per week. Ultra advertises 10× weekly usage, with no five-hour window only for verified accounts. Tax, regional and mobile-store prices can differ. Official price differences here explain the decision method; check this site’s actual prices and delivery terms on its live purchase page.
Official Claude prices and plan comparison · The Max plan · Ultra and automatic verification
Plan names, model names and account states are separate. Assess capacity benefits in models and surfaces you can access. An Ultra label does not replace inspecting Current session and Weekly. Nor can the old 20x label simply be divided by the current 10× weekly figure to establish a universal reduction percentage. The existing dedicated guide explains the history and limits.
2. Classify the bottleneck before considering an upgrade
Choose recent tasks that actually affected delivery and classify them below. One task can fit several categories. Address the first factor preventing acceptance, so a permissions problem does not become a purchase of more capacity. This is an editorial framework, not an official promise.
| Observation | Check first | Upgrade assessment |
|---|---|---|
| Repeated rework before a usage limit | Acceptance rules, inputs, model and effort | Improve configuration, then test for a remaining capability gap |
| Current session exhausted while Weekly remains | Five-hour reset and Ultra verification state | Verified Ultra may relieve the short-window interruption |
| Weekly also regularly exhausted | Task volume, model sublimits and shared surfaces | Assess the 10× weekly allowance and total extra-usage cost |
| A feature, model or organizational resource is missing | Region, rollout, roles and administrator policy | Capacity cannot replace access eligibility |
| Several people need managed access to shared material | Member identities, collaboration and organizational controls | Compare Team or Enterprise administration |
The five-hour window starts with the first message; weekly limits reset at a fixed time assigned to the account. Claude, Claude Code and Desktop count toward the same limit. One visible chat submission may trigger several tool steps. Combine surfaces when recording capacity.
3. For quality problems, test model and effort before raising the plan
Verified 5.5 publication dates are September 22 for Opus, September 28 for Sonnet and October 7 for Haiku. Anthropic positions Haiku for well-defined, high-volume and latency-sensitive tasks, Sonnet for everyday code, documents and data, and Opus for complex work needing sustained judgment. These are vendor descriptions, not this site’s measurements of your workload.
Haiku 5.5 announcement · Sonnet 5.5 announcement · Opus 5.5 announcement
In Claude, click the model name beside the send button, then choose Effort. Try lower effort for simple organization work; preserve enough reasoning for complex judgments and check acceptance. Higher effort generally uses more tokens and reaches limits sooner. Current consumer interfaces do not allow thinking to be disabled on 5.5 models. Adjust effort instead of copying API parameter combinations.
The official model-selection guide supports efficiency-first testing with upgrades for specific gaps, and capability-first testing followed by cost reduction. Apply these routes to fixed tasks rather than using max everywhere. After lowering effort for multi-step or tightly checked work, inspect whether essential steps were skipped.
Official model-selection methods · Optimize cost per completed task
4. Record the work actually lost in a sanitized log
Record representative tasks: everyday short work, one long task and work that often needs correction. Keep acceptance rules, input versions and model configurations comparable. This blank template contains no real account data. Leave missing fields unknown instead of guessing.
task_id,task_type,model,effort,entry,accepted,rework_count,blocked_by,active_wait_minutes,session_remaining,weekly_remaining
{task_id},{task_type},{model},{effort},{entry},{accepted},{rework_count},{blocked_by},{active_wait_minutes},{session_remaining},{weekly_remaining}For active_wait_minutes, count only waiting that genuinely prevents delivery and cannot be used for other work. Ten minutes of background execution is not necessarily ten minutes of lost labor. Define accepted through advance rules such as traceable citations, consistent table totals or authorized changes, rather than an impression of greater intelligence. Also keep task start and end times and a short reason why waiting could not be used for other work. Mark parallel sessions. Do not extrapolate a single good result or progress percentage into a whole month of capacity.
Also record reset times and model sublimits. Fable on Max and Ultra can use at most 50% of the original weekly pool, sharing the total with other models. This is not an additional half-week grant. Faster depletion after frequent Fable use does not by itself establish a plan reduction.
5. Three reusable prompts tying output to acceptance
These read-only templates have not been executed. Replace placeholders with sanitized material you are authorized to provide. Allow unknown answers when inputs are missing. Compare output quality before recording time and capacity; do not ask the model to guess hidden token budgets.
Task: {task_description}
Use only the material I provide and am authorized to use. First list required inputs and acceptance conditions.
Separate established facts, inferences and unknowns. Do not invent missing sources.
Produce an auditable draft and map it to each item in {acceptance_rules}.
Analyze only. Do not modify files, send messages or perform external actions.Sanitized task log: {sanitized_log}
Classify each main bottleneck: output quality, five-hour window, weekly capacity, access permissions or another cause.
Use only the log. Do not infer verification or replace unknown usage with zero.
List improvements testable with model choice, effort or task decomposition, and the evidence still needed.Review this deliverable against the original material: {artifact}
Acceptance rules: {acceptance_rules}
For each rule, report pass, fail or insufficient evidence and locate the relevant content.
Do not assume acceptance because a higher plan or model was used. Give the smallest correction list.
Review only. Do not submit, purchase or publish on my behalf.6. Is another $100 worthwhile? Count realizable marginal value
This untested arithmetic uses only the official $100 monthly web-price difference. Assume recoverable time is valued at $30/hour and accepted quality is equivalent. Use your actual price difference and time value. This is neither an income forecast nor a savings promise.
| Illustrative case | Monthly recoverable-time value | After the $100 difference |
|---|---|---|
| Save 2 hours at $30/hour | 2 × 30 = $60 | 60 − 100 = −$40 |
| Save 4 hours at $30/hour | 4 × 30 = $120 | 120 − 100 = $20 |
| No reduction in recoverable waiting | 0 × 30 = $0 | 0 − 100 = −$100 |
Do not count all screen waiting as value or automatically subtract API credit face value from the subscription fee. If quality falls, rework increases or weekly limits still interrupt work, the time-saving assumption fails. Track total cost per accepted task together with quality, manual correction and interruption counts.
7. Reconcile API grants, interactive Code and subscription windows separately
Max grants $100 and Ultra $200 in monthly API credits. Eligibility and linking conditions apply, with credits sent to one Console organization. They fund covered API applications and agents, without extending Claude, Code or Cowork subscription windows or covering those products’ extra usage.
| Surface or funds | What to inspect | Purchase assumption to avoid |
|---|---|---|
| Claude / interactive Claude Code / Cowork | Subscription windows, weekly pool and extra-usage settings | An API grant enables unlimited continuation |
| API or covered self-built agents in the linked organization | Actual API costs, balance, expiration and budgets | Credit face value is a cash-redeemable discount |
| Self-run claude -p with the linked organization API key | Authentication and initiation meet the official exception | Every IDE, desktop or Action run with -p is covered |
| Team monthly API pool | Seat-based sum, capped at $500 for the whole team | Every member gets $500 or subscription limits are pooled |
Extra credits can replace expenditure only when you already need covered API work and actually use them in that period. Unused grants expire without rollover. Use the existing tutorial to check claiming eligibility, receiving organization and actual billing. An upgrade button does not establish successful claiming.
8. Some purchase questions concern access and administration
Ultra verification is automatic, using factors including account and payment history. A temporary limit applies beforehand; checkout phone verification is separate. No deterministic threshold or deadline is published. Understand the rule before buying, then inspect Settings > Usage afterward. No particular account can be promised an immediate five-hour exemption.
Team provides member identities, collaboration and centralized administration and requires at least 2 members. Members’ subscription allowances are independent and non-transferable. Compare individual windows first for one person’s long tasks; compare seats and controls for organizational management.
Recent features show that higher prices do not grant every qualification. Personal Finance beta, announced November 5, is limited to US Pro, Max and Ultra and is unavailable on Team or Enterprise. It reads accounts but cannot transfer or invest money. The enterprise security center, centralized management and other controls announced that day have Enterprise and beta boundaries. A personal upgrade does not grant those organizational controls.
Personal Finance availability · November 5 enterprise security update
9. Define a verifiable upgrade checkpoint
Write down one problem, such as an afternoon long task blocked by a short window while weekly allowance remains. Fix material, task version and acceptance rules. Record manual corrections, failed retries and waiting separately. Parallel activity on other surfaces affects shared limits and belongs in the log too. Mark samples where the model changes, a large file is added or acceptance rules change mid-task, rather than combining them with the original configuration. Failed samples matter too: requests whose output fails acceptance still consume resources. Do not count only the request costs of successful work.
Try controlled configuration changes within the existing plan first. Before upgrading, review applicable payment, renewal and cancellation terms and stay within your budget. Keep definitions consistent afterward. A few successes do not establish unlimited long-term capacity; a few failures do not alone prove a ban or hidden reduction. Observe several task types under comparable conditions instead of deliberately maximizing consumption to discover hidden caps. Pause comparisons for access, connection or service incidents, retaining error types and times. Those interruptions have different explanations from insufficient plan capacity; do not count them as upgrade benefits without checking.
Stop expanding a trial when daily needs are met, required access is still missing, rework cancels savings or extra spending exceeds your budget. This is editorial guidance. No subscription was purchased, verification requested or task executed for the reader.
10. Upgrade decision flow
Original decision flow: each step needs actual evidence. Verify unknown states first.
- Define a real deliverable and acceptance rules
- Record model/effort, surface and window interruptions
- For quality gaps, adjust inputs and model configuration first
- For short-window gaps, check verified Ultra conditions
- For weekly or access gaps, check budgets and eligibility separately
- Compare recoverable value and live purchase terms, then decide
11. Keep five verifiable records before ordering
- Actual plan and account eligibility, beyond the product title.
- Main tasks, model and effort, with explicit acceptance rules.
- Whether Current session, Weekly, model sublimits or permissions cause interruptions.
- Covered API work, linked organization, expiration and extra-spending settings.
- Current purchase-page price, delivery scope, renewal and support terms.
This article uses official Anthropic rules to explain upgrading. It does not establish that this site’s products have the same delivery arrangement as official subscriptions. Check stock, price, account state and service terms on the purchase page and order. Nothing promises availability, official authorization, successful verification or guaranteed feature access.