Buy Pro, and buy it annually. Claude’s ladder has no volume discount anywhere: Pro at $20 buys 1x of Pro’s weekly usage, Max at $100 buys 5x and Ultra at $200 buys 10x — exactly $20 per 1x at every rung. The only discount Anthropic sells is annual billing, and it is offered on Pro and Team but not on Max or Ultra, so the cheapest usage per dollar on the price list is the lowest paid tier. Two things invert that: the monthly API credit, worth 1.00x the plan price on Max and Ultra and nothing at all on Pro, and the training-data default, which only Team and Enterprise set to none.
The ladder as Anthropic publishes it
| Plan | Monthly | Annual (per month) | Billing cycle | Weekly usage | 5-hour limit |
|---|---|---|---|---|---|
| Free | $0 | — | n/a | Limited | Limited |
| Pro | $20 | $17 ($200 up front, $16.67 effective) | Monthly and annual | 1x | 1x |
| Max | $100 | — | Monthly only | 5x | 5x |
| Ultra | $200 | — | Monthly only | 10x | None‡ |
| Team Standard | $25/seat | $20/seat | Monthly and annual | More than Pro | — |
| Team Premium | $125/seat | $100/seat | Monthly and annual | 5x a Standard seat | — |
| Enterprise | — | $20/seat | Annual, 20-seat minimum | Usage at API rates | — |
Prices and multiples verified 1 November 2026 against claude.com/pricing, the Max plan article and the Ultra plan article. Team is 2 to 150 people and the two seat types can be mixed. ‡Ultra’s “No 5-hour limit” is footnoted on Anthropic’s own page as applying to verified accounts only. Our pricing tracker carries the dated change log.
Every rung costs the same, and the only discount is at the bottom
Divide price by published multiple and the ladder’s architecture appears. This is the table that is on no Anthropic page.
| Plan | Price | Weekly multiple | Cost per 1x of Pro’s weekly usage |
|---|---|---|---|
| Pro, annual | $17 published / $16.67 effective | 1x | $16.67 |
| Pro, monthly | $20 | 1x | $20.00 |
| Max | $100 | 5x | $20.00 |
| Ultra | $200 | 10x | $20.00 |
There is no volume discount on Claude, and there is a volume penalty. Every rung is $20 per 1x, so no upgrade is ever a value decision — it is only ever a capacity decision. And because annual billing exists on Pro but not on Max or Ultra, climbing from Pro-annual to Max costs 20% more per unit of usage than staying put. Anthropic is explicit that there is no way around this: its support team “cannot issue one-off discounts or coupons”, and nonprofit discounts exist only on Team and Enterprise.
This was not always true. Until 30 October 2026 the top rung was called Max 20x and bought 20x Pro’s usage for $200 — $10 per 1x, the one genuine bulk rate on the list. Anthropic renamed it Ultra and published 10x instead of 20x at the same price, calling it a rename “with no price change”. What buyers got in exchange was the removal of the five-hour window. What they lost was the only discount on the ladder.
Max or Ultra: which limit binds you
Because the per-unit cost is identical, the Max-versus-Ultra decision is entirely structural. The two plans are constrained differently:
- Max is capped twice — “five times the Pro plan’s usage every five hours and every week”.
- Ultra is capped once — “10 times the Pro plan’s usage every week”, with no five-hour limit.
So Ultra is strictly better than Max on both axes: double the weekly ceiling and no session cap, for double the money. If your work arrives in long uninterrupted sittings, the window removal is the thing you are actually buying, and Max cannot sell it to you at any price. If your usage is evenly spread, you are paying 2x for exactly 2x and there is no structural reason to prefer either.
One risk is specific to Ultra and belongs in the purchase decision rather than the footnotes. The no-five-hour-limit benefit is conditional on verification Anthropic performs on its own schedule — “we automatically verify Ultra accounts using factors like your account and payment history”, and until then “a temporary limit applies”. There is no application and no timeline. A new Ultra subscriber can therefore pay $200 and briefly be worse off than Max on the session axis while paying double for it. Check Settings → Usage: no “Current session” bar means the account is verified.
Both plans are monthly-only, which cuts the other way and makes this cheap to get wrong. There is no annual commitment to forgo and no lock-in, and Anthropic prorates the step up: moving Max to Ultra charges “on a prorated basis for the remainder of your billing cycle” and issues prorated credits immediately. Buy one month, measure, decide.
The credit that inverts the ladder
One budget sits outside the table above because it is not plan capacity. Every paid plan except Pro and Enterprise returns a monthly Claude API credit.
| Plan | Price | Monthly API credit | Return | Net cost |
|---|---|---|---|---|
| Pro | $20 | None | 0.00x | $20 |
| Max | $100 | $100 | 1.00x | $0 |
| Ultra | $200 | $200 | 1.00x | $0 |
| Team Standard | $20/seat | $20/seat, pooled | 1.00x | $0 |
| Team Premium | $100/seat | $100/seat, pooled | 1.00x | $0 |
| Enterprise | $20/seat | None | 0.00x | $20 |
Credits verified 1 November 2026 against Anthropic’s monthly API credits article. Ratios use annual seat prices for Team; on monthly billing the credit is flat while the seat price rises, so Team’s return falls to 0.80x.
For a buyer who already spends $100 a month on the Claude API, this inverts the universal “start with Pro” advice: Max costs $0 net and carries five times the usage, while Pro costs $20 net of nothing. The cheaper plan is the more expensive one.
Four restrictions decide whether that applies to you, and they are all in Anthropic’s own words. The credits “don’t change your usage limits in Claude, Claude Code, or Claude Cowork” — they buy API calls, not chat or agent headroom. They do not cover interactive Claude Code in the terminal, IDE, desktop or web, nor Claude on Bedrock, Vertex AI or Microsoft Foundry, which rules out every EU-resident route. They need seven days on the plan, expire each billing cycle with no rollover, and link to exactly one Console organisation you cannot change yourself. To a buyer with no API workload they are worth $0.
Claude Code buyers should note one sharp distinction in the same article: claude -p run with an API key is covered and “billed as Agent SDK usage”, but run signed in to your plan it “still draws from your plan’s usage limits”. GitHub Action, IDE extension and desktop runs “count as Claude Code usage, so API credits don’t cover them, even with -p.”
The entitlement no individual plan sells: training set to none
The most consequential row on Anthropic’s comparison table is not a price. Under Model training, Free, Pro, Max and Ultra all read “Opt-out” — you are in by default and must act to leave. Team and Enterprise read “None by default.”
A $200 Ultra subscriber is on opt-out training. A $20-per-seat Team Standard subscriber is not. Team is the only self-serve way to buy a no-training default, and it costs $3.33 per seat per month more than Pro — a difference the $20 seat credit more than covers. Against two Pro-annual subscriptions at $33.34 a month, two Team Standard seats cost $40 and return $40 in pooled credits, carry more usage than Pro, and change the training default. The barrier is structural rather than financial: Team’s minimum is two seats, so the floor is $40 a month and a solo buyer cannot reach it.
If you do grow, the pooled credit is regressive. Credits are $20 per Standard seat and $100 per Premium seat, capped at $500 a month — Anthropic’s own worked example gives three Standard and two Premium seats $260. The return therefore peaks at exactly $500 of monthly seat spend (25 Standard seats, five Premium, or any mix summing to $500) and dilutes beyond it: ten Premium seats return 0.50x, twenty return 0.25x. The seat price is flat and the credit behind it is not. The GitHub Copilot ladder has the same regressive shape at a different cap, and Google’s Gemini ladder is U-shaped — a bulk rate at both ends and a premium in the middle, which is the opposite of Claude’s flat $20 per 1x.
What you cannot compute, and why it matters
Every usage figure on Claude’s price list is a multiple of Pro, and Pro’s own allowance is never published in messages, tokens or hours. Anthropic’s footnote explains why that would be hard — “How fast you reach a limit depends on the model you use, how long your chats are, and how large your files are” — but the consequence for a buyer is concrete: no Claude plan can be sized against a workload in advance, only discovered in use.
It also means a multiple can be revised while the price stays still, which is exactly what happened to the $200 tier on 30 October. Treat any allowance expressed as a multiple of another plan as a number that can move, buy monthly where monthly is offered, and re-check the allowance rather than the price. Claude Code’s weekly limits were separately cut 17% in September 2026 against a base Anthropic has never published.
Two entitlements worth checking before you pick
Fable. Anthropic’s most expensive model is not uniformly included. Pro gets it “paid with usage credits” — real extra money. Max and Ultra get it at 50% of weekly limits, with Ultra carrying a separate weekly Fable limit on top. On Team it is Premium seats only. If Fable is the reason you are buying, Pro is not the plan.
Claude Code. Included on every paid plan and on none of Free. Team Standard at $20 a seat is the cheapest route to it, which is widely mis-stated. Engineers weighing the agent rather than the chatbot should read the Claude Code review and the coding-tool cost comparison; Haiku 5.5’s 100,000-token rate cliff moves an API bill further than any plan choice.
So which one
- Free — evaluation only. Usage is “Limited” with no number attached, Projects cap at five, and Claude Code is not included.
- Pro, $17 annual / $20 monthly — the default, and the cheapest usage per dollar on the ladder. Correct for almost everyone, and the right place to stop. The full product assessment is in the Claude review; if you are still choosing a vendor, Claude vs ChatGPT and the best AI chatbots guide come first.
- Max, $100 — when you hit Pro’s caps, or when you already spend $100 a month on the Claude API and it becomes free. No discount for climbing.
- Ultra, $200 — when the five-hour window is what stops you, not when you want a better rate. There isn’t one. Verify your account status before relying on the headline benefit.
- Team Standard, $20/seat annual — two or more people, or anyone who wants training set to none by default. Strictly better than Pro on usage, credits and privacy for $3.33 a seat, if you can clear the two-seat minimum.
- Team Premium, $100/seat annual — 5x a Standard seat, and the only Team seat that includes Fable. Keep total seat spend near $500 to hold the credit at 1.00x.
- Enterprise, $20/seat — SSO, audit and 20 seats minimum, with usage billed at API rates and no credit allowance at all. Buy it for governance, not capacity.
Price a specific API workload in the API cost calculator, and check current figures for every vendor in the pricing tracker.
What changed
- 1 Nov 2026: First published. All seven plan prices, both usage multiples, every API credit allowance and the model-training defaults verified against claude.com/pricing and Anthropic’s Max, Ultra and API-credit articles. Supersedes the planned “migrate off Max 20x” guide: Max 20x was renamed, not retired, so there is nothing to migrate off — the decision it raised is the Max-versus-Ultra one answered above.
Frequently asked questions
Which Claude plan should I buy?
Pro at $20 a month, or $17 billed annually ($200 up front, which is $16.67 a month exactly), for almost everyone — and it is the cheapest usage per dollar on the entire ladder, because annual billing is offered on Pro and Team but not on Max or Ultra. Move up only when you are actually hitting caps, because no upgrade carries a volume discount: Pro buys 1x of Pro's weekly usage for $20, Max buys 5x for $100 and Ultra buys 10x for $200, which is exactly $20 per 1x at every rung. Two facts change that answer. If you already spend $100 or more a month on the Claude API, Max costs $0 net of its $100 monthly API credit while Pro costs $20 net of nothing, so the cheaper plan is the more expensive one. And if you want Anthropic's training default set to none rather than opt-out, Team is the only self-serve way to buy it, from $20 per seat per month on annual billing with a two-seat minimum.
Is Claude Max worth it, or should I go straight to Ultra?
Neither is better value than the other — they cost the same $20 per 1x of weekly usage — so the question is which limit binds you. Max is constrained twice: five times Pro's usage every five hours and five times Pro's usage every week. Ultra is constrained once: ten times Pro's usage every week, with no five-hour limit at all. Ultra is therefore strictly better than Max on both axes, double the weekly ceiling and no session cap, for double the money. If your work arrives in long uninterrupted sessions, the removal of the five-hour window is the thing you are buying and Max cannot sell it to you at any price. If your usage is spread evenly, you are paying 2x for exactly 2x and the choice is pure capacity. One purchase risk is specific to Ultra: the no-five-hour-limit benefit applies only to accounts Anthropic has automatically verified, so a new subscriber can pay $200 and still see a session cap until that happens.
What is a verified Ultra account, and can I make my account one?
No, and this is the part to read before paying. Anthropic states that to prevent fraud and abuse it 'automatically verifies Ultra accounts using factors like your account and payment history', that 'Verified accounts don't have a five-hour limit', and that 'If your account isn't verified yet, a temporary limit applies.' There is no application, no published criteria beyond account and payment history, and no stated timeline — only 'We remove it automatically once your account is verified, so there's nothing you need to do.' It also pre-empts the obvious workaround: 'Verifying your phone number at checkout is a separate step and doesn't remove this limit.' To check where you stand, Anthropic says to open Settings then Usage — if there is no Current session bar, the account is verified. So Ultra's headline advantage over Max is conditional on a decision the buyer cannot influence and may not have received at the moment of purchase.
Do the monthly API credits make Max or Ultra free?
Net of the credit, yes, and that genuinely inverts the ladder — but only for a buyer who already has API spend. Max includes $100 a month in Claude API credits against a $100 price and Ultra includes $200 against $200, both exactly 1.00x, while Pro includes nothing and Enterprise includes nothing. So for anyone already spending $100 a month on the Claude API, stepping Pro to Max takes net outlay from $20 to $0 and multiplies the usage allowance by five. The restrictions are what make this conditional rather than universal. Anthropic states plainly that 'The credits don't change your usage limits in Claude, Claude Code, or Claude Cowork' — they buy API calls, not chat or agent headroom. They do not cover interactive Claude Code in the terminal, IDE, desktop or web, nor Claude on Bedrock, Vertex AI or Microsoft Foundry. They require seven days on the plan, expire at the end of each billing cycle with no rollover, and link to exactly one Console organisation that you cannot change yourself. To a buyer with no API workload they are worth $0 and Max is a $100 chat subscription.
Is Claude Team cheaper than buying individual Pro seats?
It is more expensive per seat and better value anyway, which is unusual on this price list. Team Standard is $20 per seat per month on annual billing against Pro's $16.67 effective, so $3.33 per seat more. For that you get more usage than Pro, $20 per seat per month in pooled API credits that more than cover the difference, and the one entitlement no individual plan sells at any price: Anthropic's pricing page sets model training to 'None by default' on Team and Enterprise, against 'Opt-out' on Free, Pro, Max and Ultra. The catch is the two-seat minimum, so the floor is $40 a month and a solo buyer cannot reach it. Watch the pooled credit cap if you grow: credits are $20 per Standard seat and $100 per Premium seat but the pool is capped at $500 a month, so the return peaks at exactly $500 of monthly seat spend — 25 Standard seats, or five Premium seats, or any mix adding to $500 — and every seat beyond that dilutes it. Ten Premium seats return 0.50x, twenty return 0.25x. The seat price is flat; the credit behind it is regressive.
Why can't I work out how much usage a Claude plan actually gives me?
Because Anthropic does not publish the base figure, and this is the most important limitation on the whole price list. Every usage number on claude.com/pricing is expressed as a multiple of the Pro plan — Free 'Limited', Pro '1x', Max '5x', Ultra '10x' weekly — and Pro's own allowance is never stated in messages, tokens or hours. The footnote explains why it would be hard to: 'How fast you reach a limit depends on the model you use, how long your chats are, and how large your files are.' The practical consequence is that no Claude plan can be budgeted against a workload in advance, only discovered in use, and that a multiple can be revised while the price stays still — which is exactly what happened on 30 October 2026, when the $200 tier's published weekly multiple went from the 20x in its old name to 10x. Buy monthly where monthly is offered, measure a real week, then commit.