Claude Usage Limits Explained: Max and Ultra, Team and Enterprise, Monthly API Credits and Costs

1. What changed from 5x and 20x: check the measurement window first

The former Max 5x plan is now Max, and the former Max 20x is Ultra. The official help page says existing 20x subscribers move to Ultra automatically, with the web subscription price remaining $200 a month. Max costs $100 a month on the web. Official snapshots retained by this site on October 31, 2026 already contained these names and usage rules. They establish what was publicly visible then, rather than a precise effective time for every policy.

The most easily misread change is “20x became 10x.” A readable October 23, 2026 archive of the official Max help page explicitly defined 20x as Pro usage per session. Current Ultra documentation specifies 10× Pro usage per week. The denominators cover different periods, so dividing the figures does not establish that weekly capacity was halved. Comparable historical budgets are also insufficient to prove that every account’s actual capacity remained completely unchanged.

Historical wording, current published rules and the state displayed in an individual account are three different kinds of evidence. When reading an older announcement or a seller’s description, first identify whether a multiplier refers to each five-hour window, a week or a dollar-denominated API balance. Comparisons need consistent models, windows, input/output structures and periods. This article has not measured any account’s hidden allowance.

Claude: current plan prices and usage comparison · Claude Help: Max and the former Max 5x plan · Claude Help: Ultra and the former Max 20x plan · Official Max help page: archive captured October 23, 2026

2. Current individual plans: read five-hour and weekly limits separately

The following are US website subscription prices in US dollars, excluding applicable taxes. Regional and mobile-store prices can differ. Annual Pro is a single $200 payment; the website’s approximately $17 per month is an annual equivalent. Max and Ultra currently renew monthly. Multipliers reflect the published allowance, rather than a guaranteed number of messages.

Claude: current plan prices and usage comparison · Claude Help: Max and the former Max 5x plan · Claude Help: Ultra and the former Max 20x plan

3. Why work can stop despite remaining allowance: windows, models and context

When both five-hour and weekly limits apply, reaching either can prevent further included usage. Check the account’s reset times in Settings > Usage. Each five-hour window is not a separate grant that can be accumulated or carried forward. Removing Ultra’s short usage window does not remove its weekly cap.

Consumption depends on the model, reasoning effort, conversation length, attachments and multi-step tool execution. A request to “finish changing this project” can lead to repeated reasoning and tool-result exchanges. One user message in the interface does not represent a fixed cost. Claude and Claude Code subscription usage should not be treated as two independent allowances.

Max and Ultra currently have a separate weekly Fable limit, listed as 50% of weekly usage in the website comparison. Fable is constrained by both the overall weekly pool and its model-specific cap; this is not an additional 50% allowance. Pro requires usage credits for Fable. Running out of context is a length constraint, and waiting for a usage reset will not enlarge a single conversation’s context window.

Claude: current plan prices and usage comparison · Claude Help: Ultra and the former Max 20x plan · Claude Help: usage limits and context length · Claude Help: Claude Code with Team or Enterprise · Claude Help: Fable allowances by plan

4. Ultra’s five-hour exemption is conditional: how to check your status

The official explanation says Ultra accounts are automatically verified using factors such as account and payment history. The five-hour limit is removed once verification is complete; a temporary limit applies beforehand. The help page publishes no deterministic scoring formula, minimum account age, payment amount or completion deadline.

The practical check is Settings > Usage. The official indicator of verification is that the Current session progress bar no longer appears. Verifying a phone number at checkout is a separate step and does not directly remove this limit. General identity verification or CVP approval should not be treated as proof that Ultra status has changed either.

For troubleshooting, record the plan name, the Current session and Weekly displays, and their reset times before contacting official support. A successful payment, a product label or someone else’s screenshot cannot establish your own weekly pool or verification state.

Claude Help: Ultra and the former Max 20x plan

5. Team: included usage belongs to each member; API grants form a team pool

Team currently serves organizations of 2 to 150 people and supports a mix of Standard and Premium seats. US web pricing for Standard is $25 per person monthly, or $20 per person per month with annual billing. Premium is $125 or $100 respectively. Annual comparisons must use the number of paid seats; the price of one Team seat is not the price for an entire organization.

The official explanation gives both seat types five-hour and cross-model weekly limits. Standard offers approximately 1.25× Pro usage per session, and Premium approximately 6.25×. These are per-session figures and must not be copied into a weekly comparison. The current material does not publish a fixed weekly token budget that can be promised.

Members cannot transfer or combine their included subscription allowances. When one person reaches a limit, another person’s unused allowance does not automatically move to them. The monthly API pool discussed later is a separate organizational balance. Subscription charges for added or upgraded seats are prorated over the remaining billing period; the timing of the next API grant must be checked separately.

Claude Help: the Team plan · Claude Help: Team billing · Claude Help: monthly API credits for Max, Ultra and Team · Claude Help: Claude Code with Team or Enterprise

6. What “Business” means, and how Enterprise is billed

As of the verification date, Claude’s public pricing lists Team and Enterprise as its organizational plans, without a separate plan named Business. “Business” may be a generic description of commercial use or a seller-defined label. Assess limits using the actual plan, seat type and contract on the official bill, rather than importing another provider’s Business rules.

The published offer for new Enterprise purchases is $20 per seat per month, billed annually, plus usage metered at API rates. The seat fee does not include that usage. Self-serve purchases require at least 20 seats, while the official sales-assisted explanation lists a 50-seat minimum. Existing and special contracts need their own review. Under the published self-serve offer, 20 base seats cost $4,800 a year before any usage.

This consumption-based Enterprise arrangement does not use Team’s individual five-hour and weekly pools. Organizational, member or group budgets and actual consumption still constrain service. Exhausting a self-serve prepaid balance can stop usage across the organization, and sales-assisted billing does not imply the absence of budget limits. Legacy seat-based Enterprise and newer consumption-based contracts must not be conflated.

“Billed at API rates” describes a pricing basis; it does not automatically grant a Console API credit pool. The current monthly API credit program for Max, Ultra and Team explicitly excludes Enterprise. Check API integration access, organizational subscription consumption and any separate API contract individually.

Claude: current plan prices and usage comparison · Claude Help: the Enterprise plan · Claude Help: Enterprise billing · Claude Help: Enterprise consumption guide · Claude Help: monthly API credits for Max, Ultra and Team · Claude Help: usage credits for Team and seat-based Enterprise

7. Monthly API grants by plan: check the amount and who shares it

These monthly API credits enter a Claude Console organization after eligibility and linking requirements are met. They are denominated in dollars and are separate from the percentage meters shown for subscription usage. Free, Pro and Enterprise are outside this particular monthly grant program; this does not mean they can never qualify for another trial or targeted promotion.

Claude: current plan prices and usage comparison · Claude Help: monthly API credits for Max, Ultra and Team · Claude API: subscriber credits and restrictions

8. Claiming credits: link the right Console organization after seven days

An eligible subscription must be active and in good standing. New subscribers can claim after seven days on the eligible plan. The subscription may have been purchased on the web or mobile, but claiming takes place on the claude.ai website. You do not need to add a payment card to Claude Platform first.

A Max or Ultra subscriber opens Settings > Billing. For Team, a Primary Owner or Owner opens Organization settings > Billing. In the API credits section, select Link organization, choose an existing Console organization or create one, review the terms and confirm. On the Console side, the necessary role is Owner, Admin or Billing.

Only one Console organization can be linked, and a Console organization can receive this grant from only one plan. Multiple personal subscriptions cannot simply be stacked into the same API wallet. You cannot change the linked organization yourself afterward; changes require support. Confirm its name, membership and intended use before proceeding, to avoid sending credits to a disposable test organization or somebody else’s project.

After claiming, check the Promotional credits amount and expiration date at platform.claude.com/settings/billing. If the option is missing after seven days, first check the plan, roles and any current official phased-availability guidance, then contact support. Seven days is an eligibility requirement for claiming credits, not a deadline for Ultra’s automatic verification.

Claude Help: monthly API credits for Max, Ultra and Team · Claude API: subscriber credits and restrictions

9. Team pooling, upgrades and expiration: grants are not a rollover deposit

The Team monthly grant is min(20 × Standard seats + 100 × Premium seats, 500) dollars. Three Standard seats plus two Premium seats receive $260. Ten Premium seats still receive $500, rather than $1,000. Buying more seats cannot exceed this program’s team-wide cap.

Credits refresh each billing period, with monthly grants even on annual subscriptions. Unused credits expire at the end of their period and do not roll over. They are used before purchased API credits. API keys and workspaces within the same linked organization all consume this one pool.

Team grants use the seats applicable when the grant is issued. Platform documentation says added seats enter the next billing period’s grant. Do not assume that prorated seat charges also produce an immediate prorated API grant. Help and platform pages use slightly different wording for the calculation time; check actual posted credits or support for the precise amount at a boundary.

The Max-to-Ultra upgrade documentation specifies a prorated increase in API credits, followed by $200 each billing period. Credits for the Google Play purchase path may arrive later, no later than the next renewal. Canceling or moving to an ineligible plan stops future grants, while issued credits retain their existing expiration rules.

Claude Help: monthly API credits for Max, Ultra and Team · Claude API: subscriber credits and restrictions · Anthropic: Supplemental Credit Terms

10. Which activity the grant covers, and which is billed separately

Using the same Claude model does not imply using the same billing product. Identify the surface, authentication and how execution was initiated before deciding whether monthly API credits cover it. Model availability, additional eligibility and service support still apply; a balance does not automatically remove access conditions.

Claude Help: monthly API credits for Max, Ultra and Team · Claude API: subscriber credits and restrictions · Claude Help: Claude Code with Team or Enterprise · Claude API: Fable 5.1 pricing and availability

11. Route work between four balances: identify the surface before the funds

At least four independent concepts matter in practice: included subscription usage windows, the subscription-side extra usage budget, monthly API credits and purchased API credits. The latter two cannot directly replenish the first two. You can move work to a different surface after a limit is reached, but must check authentication, data processing and cost ownership again.

When monthly API credits run out, an organization with purchased funds or configured auto-reload can continue using those funds. With no other credits, API usage stops until new credits arrive or are purchased. Charges do not automatically move to the personal Claude subscription bill. Sales-invoiced organizations settle overages using their existing billing arrangement.

One easily missed detail is that platform documentation says auto-reload triggers from the purchased balance. A low purchased balance can therefore trigger a recharge even while monthly credits remain. To control extra spending, review both the organization’s auto-reload settings and spending limits, rather than watching only the promotional balance.

Claude Help: monthly API credits for Max, Ultra and Team · Claude API: subscriber credits and restrictions · Claude Help: usage credits for Team and seat-based Enterprise

12. What API usage costs: model rates and long-prompt tiers

This table uses standard global service prices for the first-party Claude API, in US dollars per million tokens (MTok). It excludes taxes, tools, runtime, caching, Batch and other price modifiers. Caching and regional pricing are considered separately below. The table supports estimates and does not establish that an account has access to a model.

Haiku 5.5 requires particular attention to input-length tiers. Input and output rates differ for prompts at or below 100,000 tokens and those above the threshold. Cache reads and writes also count when determining prompt length; looking only at uncached input_tokens is insufficient.

Fable API prices and subscription Fable weekly caps are different concepts. Monthly API credits are dollar balances depleted at applicable API rates. The individual subscription rule allowing Fable to use up to half a weekly pool cannot be converted into a fixed API dollar allowance.

Claude API: current model and feature pricing · Claude API: prompt caching prices and usage fields · Claude API: Fable 5.1 pricing and availability · Claude API: Sonnet 5.5 pricing · Claude API: Opus 5.5 pricing · Claude API: Haiku 5.5 prices and prompt-length tiers

13. How far $100 or $200 could go: a fixed-workload calculation

The following is an original arithmetic illustration, not an API experiment. Assume each request has exactly 10,000 billable uncached input tokens and 2,000 billable output tokens, including any thinking. It uses no tools, caching or other billable features, and the global standard rates above. Haiku uses the tier below 100,000 prompt tokens.

Cost per request = input tokens ÷ 1,000,000 × input rate + output tokens ÷ 1,000,000 × output rate. The theoretical count is floor(available dollar balance ÷ cost per request). It describes only these artificially fixed assumptions. Actual tasks vary with output, retries, multiple calls, model availability and rate limits.

A more useful purchasing measure is the actual cost of an accepted task. A task might make ten requests and ingest many tool results, or succeed on its first attempt. Repeated rework can make a cheaper model more expensive overall. This table cannot be used to infer subscription message counts or weekly allowances.

Claude API: current model and feature pricing · Claude API: Fable 5.1 pricing and availability · Claude API: thinking tokens and billable output · Claude API: Sonnet 5.5 pricing · Claude API: Opus 5.5 pricing · Claude API: Haiku 5.5 prices and prompt-length tiers

14. Savings from caching and Batch: charge the write before the hits

Caching does not discount all inputs uniformly. A typical five-minute cache write costs 1.25× the base input rate, while a one-hour write costs 2×. Read multipliers vary by model. Current Sonnet 5.5 and Opus 5.5 reads cost 0.05× base input, and Fable 5.1 reads cost 0.025×. An older article’s universal 0.1× assumption should not be applied to every model.

The following Sonnet 5.5 example fixes a repeated 10,000-token prefix, 1,000 new input tokens and 2,000 output tokens per request. When the prefix is written, charge it only as a cache write, without charging uncached input again. On a hit, charge it only as a cache read. Assume the content meets caching requirements, later requests actually hit, and the entry has not expired.

Across ten repeated runs, no caching costs $0.42. One five-minute write followed by nine full hits costs $0.047 + 9 × $0.023 = $0.254. The approximately 39.5% saving includes output costs and is an original worked example, rather than a claim that every task receives that discount.

Message Batches offers a 50% discount on standard input and output rates for supported asynchronous requests, with caching and other combinations governed by the documentation. It has asynchronous waiting and feature-support constraints. It is not a free speed-up for real-time requests. Choose work that can tolerate batching, then validate complete responses and quality.

Claude API: current model and feature pricing · Claude API: prompt caching prices and usage fields · Claude API: batch processing

15. Why funds disappear faster than expected: five sources of extra consumption

First, context is repeatedly sent back. History, system instructions, tool definitions, retrieved material and tool results may all enter later inputs. As a conversation grows, actual input can rise even when the user adds very little text. Second, output and reasoning matter. Billing uses the service-reported output_tokens, which includes raw thinking. output_tokens_details.thinking_tokens is a subset and must not be added again. The length of the final visible summary is not a substitute.

Third, tools and agent execution can add costs. Search, code execution and Managed Agents may involve additional tool or runtime fees, while tool results add tokens. Calculating only one text exchange misses this. Fourth, model and pricing modifiers matter, including Haiku crossing a long-prompt tier, an applicable US-only inference premium or another enabled service tier.

Fifth, retries and parallel execution matter. Multiple agents may each load context, and blindly rerunning after a disconnect can create additional calls. Set request-count, concurrency and total-budget bounds for each task, and record complete terminal states. The number of visible “submissions” alone is not a cost ledger.

Claude API: current model and feature pricing · Claude API: prompt caching prices and usage fields · Claude API: Usage and Cost Analytics · Claude API: thinking tokens and billable output

16. A positive balance can still be blocked: four limits and two reset clocks

API balances, the organization’s monthly spend cap, workspace limits, and request/input/output token rates are different controls. Monthly API grant consumption also counts toward the organization’s monthly spend cap. Platform documentation says linking an eligible subscription grants at least Start, while a monthly grant strictly greater than $200 grants at least Build. Ultra’s $200 therefore does not establish Build eligibility under that rule alone.

The organization’s monthly spend cap resets by calendar month, whereas promotional credits refresh by subscription billing period. The clocks can diverge. Reaching an organizational cap is not necessarily resolved immediately by buying funds or receiving new credits. Current documentation lists monthly spend caps of $500 / $1,000 / $200,000 for Start / Build / Scale. Actual Console settings and contracts still apply.

Teams can select a project under Workspaces in Console and configure Spend limits and alerts. Organizational controls can be viewed or adjusted under Billing > Spend limits. The Default Workspace currently does not support this separate limit. A workspace without its own cap remains subject to the organizational cap. Application-side concurrency and task budgets are still needed to reduce uncertainty from parallel execution.

Classify the failure first: exhausted credits point to Billing; budget caps to organization and workspace settings; rate limits to response limit information. Do not interpret every failure as an account ban or assume auto-reload raises rates or budgets. The grant itself does not count toward the cumulative purchased spend used to qualify for a higher tier.

Claude API: subscriber credits and restrictions · Claude API: rate limits and organization/workspace spend caps · Claude API: workspace management

17. Audit actual consumption: separate usage, cost and balance

Request-level usage explains tokens. Prompt caching documentation defines input_tokens as uncached input. Total input is the sum of input_tokens, cache_creation_input_tokens and cache_read_input_tokens. Five-minute and one-hour cache-write details are components of total cache writes, so adding them again would double-count.

The synthetic usage fragment below illustrates fields and is not a real response from an executed call. At standard Sonnet 5.5 rates, 1,000 uncached input + 10,000 cache-read + 2,000 output tokens cost $0.023 in text tokens, excluding other feature fees.

For organization-wide totals, use the current Usage and Cost Analytics API or Console reporting. The older Usage and Cost API is marked legacy. The current cost endpoint’s cost.amount is a decimal string in cents, representing cost after discounts and before credit deductions. It is neither a remaining balance nor necessarily the amount already due on a bank card. Aggregate data usually updates after about four hours and can be delayed up to twenty-four hours. Reconcile it separately from Billing balances rather than treating it as a real-time stop-loss signal.

Build a reproducible small sample: record an internal task ID, actual model, organization/workspace, usage, tools/runtime, terminal status and acceptance result. Separately record grant postings, purchased funds, expiration and auto-reloads from Billing. Reconcile a few fixed tasks first, then estimate a monthly budget using the same definitions. This article has executed no paid model calls on the reader’s behalf.

{
  "input_tokens": 1000,
  "cache_creation_input_tokens": 0,
  "cache_read_input_tokens": 10000,
  "output_tokens": 2000
}

Claude API: prompt caching prices and usage fields · Claude API: Usage and Cost Analytics

18. Choose a plan around the actual bottleneck: four pre-purchase checks

For long individual tasks, first distinguish short-window limits from the weekly total. If Max repeatedly hits the five-hour limit while weekly allowance remains, verified Ultra’s lack of a five-hour window may fit the work’s rhythm better. If the weekly pool is regularly exhausted, consider the 10× weekly allowance alongside model selection, task design and extra usage cost. The old 20x name is insufficient.

For teamwork, establish seat types, member allowances and management needs before calculating the API pool. With three Standard and two Premium seats on annual-equivalent pricing, the base seat cost is $260 per month and the nominal monthly API grant is also $260. Their uses, expiration and coverage differ. The matching amounts do not make the subscription free or the credits redeemable for cash.

For a self-built application, focus on measured fixed-task costs, peak rates, shared balances and data access. The API can be used separately without an eligible subscription. Developers who already have Max or Ultra can route suitable work through the linked organization’s monthly credits, with budgets configured. Continuous interactive Code work remains governed by its own product rules.

For enterprise purchasing, document base seat charges, metered usage, budget controls and API contracts separately. If a seller says Business, request the actual official plan name. This article’s tables describe Anthropic’s official prices and rules. Delivery, prices and after-sales terms for products on this site are governed by their product and order pages.

• Check the actual plan name, seat type and purchase channel.
• Inspect Current session, Weekly, model-specific caps and reset times separately.
• Confirm the linked Console organization, credit expiration, purchased funds and auto-reload.
• Measure unit cost with your own fixed tasks and configure organization/workspace budgets before expanding usage.

Claude: current plan prices and usage comparison · Claude Help: Ultra and the former Max 20x plan · Claude Help: the Team plan · Claude Help: the Enterprise plan · Claude Help: Enterprise billing · Claude Help: monthly API credits for Max, Ultra and Team · Claude API: subscriber credits and restrictions