> ## Content Index > Fetch the complete content index at: https://www.implicator.ai/llms.txt > Use this file to discover other available public pages before exploring further. # Anthropic's Usage-Based Billing Is Exact. Its Plan Limits Are Vague by Design. - URL: https://www.implicator.ai/anthropics-usage-based-billing-is-exact-its-plan-limits-are-vague-by-design/ - Published: 2026-05-14T22:32:07.000Z - Updated: 2026-05-14T22:32:07.000Z - Description: Anthropic documents its API rates and plan prices to the dollar, then leaves the token size of every Claude Code subscription limit undefined. The vague middle is the surface it re-tunes as compute allows, and the SpaceX Colossus 1 lease showed how fast it moves. - Author: Marcus Schuler - Tags: Analysis Anthropic raised Claude Code's weekly usage limits by 50 percent on May 13, an increase scheduled to run through July 13\. The change followed a [May 6 announcement](https://www.anthropic.com/news/higher-limits-spacex?ref=implicator.ai) that doubled Claude Code's five-hour session limits and permanently removed a [weekday peak-hours throttle](https://www.implicator.ai/opinion-anthropic-knew-the-math-it-sold-the-tickets-anyway/) for Pro and Max subscribers, published alongside news that Anthropic had leased SpaceX's Colossus 1 data center. In neither announcement did the company state the token size of any subscription's usage limits. The pattern holds across Anthropic's billing. Its API pricing page lists exact per-token rates for every model, and its plan page lists exact monthly prices. Neither the help center nor the Claude Code documentation states how many tokens a Pro or Max subscription includes, and a May 14 review of the company's current billing pages found no per-plan limit table. Anthropic has revised those undisclosed limits repeatedly over the past year. Key Takeaways - Anthropic publishes exact per-token API rates and exact monthly plan prices, but never the token size of a Claude Code subscription limit. - Claude Code runs on token consumption underneath; subscribers see usage bars and Pro-multipliers instead of a denominator. - Anthropic announced its SpaceX Colossus 1 data-center lease and higher Claude Code limits in a single May 6 post. - The next checkpoint is July 13, when the 50 percent weekly-limit increase is set to expire. AI-generated summary, reviewed by an editor. [More on our AI guidelines](https://www.implicator.ai/about/). ## The published prices Anthropic's [API pricing page](https://platform.claude.com/docs/en/about-claude/pricing?ref=implicator.ai), fetched May 14, lists Claude Opus 4.7 at $5 and $25 per million input and output tokens, Sonnet 4.6 at $3 and $15, and Haiku 4.5 at $1 and $5\. Cache reads cost a tenth of the base input rate, and batch jobs run at a 50 percent discount. The page also specifies that current models "include the full 1M token context window at standard pricing," and notes that Opus 4.7's new tokenizer "may use up to 35% more tokens for the same fixed text." The subscription page is equally specific on price. Pro costs $20 a month, or $17 billed annually; Max costs $100 or $200 a month; and Team is sold by the seat with a five-seat minimum. Anthropic changes these figures rarely and announces them when it does. Every Anthropic Plan, Side by Side | Plan | Price | Claude Code | Limits and notes | | --------------------------- | --------------------------------------------------- | ---------------------------------- | -------------------------------------------------------------------------------------------------------------------------------- | | Consumer | | | | | **Free** | $0 | Chat-based coding onlyNo CLI agent | Smallest tier, cut further at peak hoursNo weekly cap. Sonnet only, with limited access to other models; no extra-usage option. | | **Pro** | $20/mo$17/mo annual, $200 upfront | Included | Five-hour session, doubled May 6Weekly cap up 50% through July 13\. Shares a usage pool with chat, historically Sonnet-weighted. | | **Max 5x** | $100/moMonthly only | Included | 5x Pro, separate Opus sub-limitFull Opus access. Shares pool with chat. | | **Max 20x** | $200/moMonthly only | Included | 20x Pro, separate Opus sub-limitFull Opus access. Shares pool with chat. | | Team | | | | | **Team Standard** | $25/seat/mo$20 annual, 5-seat minimum | IncludedSince Jan 16, 2026 | 1.25x Pro, per memberOne stale help doc still lists no Code access. | | **Team Premium** | $125/seat/mo$100 annual | Included | 6.25x Pro, per memberTwo weekly caps, all-models and Sonnet-only. Opus 4.5 by default. | | Enterprise | | | | | **Enterprise (self-serve)** | $20/seatPlus usage costs | IncludedSingle seat type | Usage-based, rolling windowBilled by consumption. | | **Enterprise (negotiated)** | CustomPress reports a \~$50k/yr floor (unconfirmed) | Premium seats only | NegotiatedStandard seats exclude Claude Code. | | API | | | | | **API (pay-as-you-go)** | Per tokenBy usage tier | Via API key in the CLI or VS Code | No subscription capPer-minute and monthly spend caps by tier. Billed per token. | Prices in USD, current as of May 14, 2026\. Anthropic revises plan terms frequently. ## The unpublished limits Claude Code's cost documentation states that the tool "charges by API token consumption," counting input, output, and cached tokens. For a developer using an API key, that consumption produces an itemized bill. For a subscriber, the same documentation says the cost figure "isn't relevant for billing purposes," and that subscribers instead "see plan usage bars and activity stats." Those bars are governed by two limits, a five-hour rolling session window and a seven-day weekly cap, and Anthropic publishes neither as a token count. Its help-center articles give the higher tiers only as multipliers of Pro, with Max 5x at five times Pro and the two Team seat types at 1.25 and 6.25 times. Anthropic has not published what a Pro plan includes in tokens or in hours. A widely repeated estimate of 40 to 80 hours of Sonnet per week on Pro comes from a July 2025 statement Anthropic gave TechCrunch, echoed in a help-center article from the same month. The figure has not been updated since, though the limits behind it have changed repeatedly. Anthropic introduced weekly caps in August 2025 and throttled weekday peak hours, between 5 and 11 a.m. Pacific, in March. It removed that throttle on May 6 and raised weekly limits again on May 13\. The help center reserves the right to "limit your usage in other ways, such as weekly and monthly caps... at our discretion." What Anthropic Publishes, and What It Doesn't | What Anthropic publishes | What it leaves undisclosed | Why the gap matters | | --------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------- | | **API token rates**Opus 4.7 at $5 and $25, Sonnet 4.6 at $3 and $15, Haiku 4.5 at $1 and $5 per million tokens. | **Nothing**API billing is specified in full. | An API customer can price a workload before running it. | | **Plan prices**Pro at $20 a month, Max at $100 or $200, Team sold by the seat. | **The size of the allowance**What a Pro, Max, or Team plan includes, in tokens or hours. | Subscribers see the monthly price, not the denominator behind the usage bar. | | **Reset clocks**A five-hour session window and a seven-day weekly cap. | **The size of either clock**How many tokens or hours each window holds. | A reset time says when access returns, not how much was bought. | | **Model fallback**Older help text cited 20% and 50% Opus switch points on Max. | **The current downgrade threshold**When Claude Code drops a session from Opus to Sonnet. | Output quality can change before any API-style bill appears. | | **Programmatic credits**June 15 pools of $20, $100, and $200, billed at API rates. | **An uncapped subscription route**Any flat-rate path for third-party agent harnesses. | OpenClaw-style use returns only as a separate metered lane. | Get Implicator.ai in your inbox Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST. Email address Subscribe Check your inbox. Click the link to confirm. No spam. Unsubscribe anytime. ## The economics underneath Claude Code is sold at a flat rate, but the compute behind it is metered and capacity-constrained. An [Implicator analysis in March](https://www.implicator.ai/claudes-rate-limits-arent-a-capacity-problem-theyre-a-math-problem/) estimated that a $20-a-month Pro subscriber running Claude Code intensively costs Anthropic close to $60 in compute, and a $200 Max subscriber close to $570\. Anthropic's own explanation for the weekly caps cited a small number of accounts that ran Claude Code continuously and consumed capacity "disproportionately." Anthropic cited the same economics when it restricted third-party tools. After agent harnesses such as OpenClaw let subscribers route API-scale workloads through a flat-rate login token, the company said it had been "effectively subsidizing API-equivalent workloads at subscription prices," and it [blocked that authentication path on April 4](https://www.implicator.ai/anthropic-cuts-openclaw-from-claude-subscriptions-citing-unsustainable-compute-costs/). It said it would restore third-party access on June 15 through a separate credit pool, capped at $20, $100, and $200 by tier and billed at full API rates. ## The Colossus 1 lease On May 6, Anthropic leased the entire Colossus 1 data center in Memphis from SpaceX. More than 220,000 NVIDIA GPUs, more than 300 megawatts, online within a month. xAI's statement said Anthropic would use the compute to "directly improve capacity for Claude Pro and Claude Max subscribers," and Anthropic published the lease and the Claude Code limit increase in a single announcement. Colossus 1 is controlled by Elon Musk. His Grok model competes with Claude, and he had called Anthropic "misanthropic" three months earlier before softening his position. New Street Research and other analysts estimate the lease at $3 billion to $6 billion a year on a reported four-year term, against an Anthropic revenue run rate that has reportedly passed $30 billion. TechCrunch and the developer Simon Willison both described the deal as an unusual supply-chain dependency for Anthropic. ## A record only Anthropic keeps Anthropic's next scheduled change is July 13, when the 50 percent weekly increase is set to expire. The company has not said whether it will let the increase lapse or fold it into the standard limit. Either way, subscribers will learn of the outcome the way they learned of every change before it, through an announcement. Those announcements describe movements in numbers Anthropic has never published. The May 6 post doubled Claude Code's five-hour limits; the May 13 update raised the weekly cap by 50 percent. The company disclosed neither limit's size before the increases, and has disclosed neither since. That leaves the announcements unverifiable by design. Anthropic can say a plan grew more generous, or quietly tightened, and a subscriber has no published figure to check it against. Frequently Asked Questions Does Claude Code run on tokens even for subscribers? Yes. Anthropic's cost documentation says Claude Code "charges by API token consumption." API-key users get an itemized bill; subscribers see plan usage bars and reset clocks instead, and Anthropic says the underlying cost figure "isn't relevant for billing purposes" for them. Does Anthropic publish the token size of Pro or Max limits? No. Anthropic publishes exact API rates and plan prices, but describes subscription tiers only as multipliers of Pro, such as Max 5x. It does not publish what a Pro plan includes in tokens or hours, and a May 14 review found no per-plan limit table. Are the "40 to 80 hours" Pro estimates still accurate? They are outdated. That range comes from a July 2025 Anthropic statement to TechCrunch and a help-center article from the same month. Anthropic has since added weekly caps, throttled and unthrottled peak hours, and raised limits twice, without republishing the figures. Why does the SpaceX deal matter for billing? Anthropic announced its lease of SpaceX's Colossus 1 data center and a Claude Code limit increase in a single May 6 post. The pairing shows subscription limits now move with compute supply, not only with pricing decisions. What happens on July 13? The 50 percent weekly-limit increase Anthropic introduced on May 13 is scheduled to expire July 13\. The company has not said whether it will let the increase lapse or fold it into the standard limit. AI-generated summary, reviewed by an editor. [More on our AI guidelines](https://www.implicator.ai/about/). [Anthropic Briefly Removes Claude Code From $20 Pro Plan, Calls It 2% TestAnthropic briefly removed Claude Code from its $20-per-month Pro plan on Tuesday afternoon, flipping the feature to unavailable across pricing pages and support documents before reversing the change wThe Implicator![](https://www.implicator.ai/content/images/2026/04/2026-04-22-12.23.57-claude_code_pro@2x.webp)](https://www.implicator.ai/anthropic-briefly-removes-claude-code-from-20-pro-plan-calls-it-2-test-2/) [OPINION: Anthropic Knew the Math. It Sold the Tickets Anyway.On March 23, Anthropic's paying Claude subscribers discovered their sessions had been throttled without notice. Max users handing over $200 a month watched their daily allowance vanish on a single proThe Implicator![](https://www.implicator.ai/content/images/2026/03/2026-03-27-17.12.15-rate_limits_ticket@2x.webp)](https://www.implicator.ai/opinion-anthropic-knew-the-math-it-sold-the-tickets-anyway/) [Claude's Rate Limits Aren't a Capacity Problem. They're a Math Problem.A developer's wife asked Claude one question about used Chevy Blazer EVs this week and immediately hit her session limit. One question. One answer. Locked out. She messaged her husband to ask if she wThe Implicator![](https://www.implicator.ai/content/images/2026/03/2026-03-27-16.15.54-claude_rate_limits_math@2x.webp)](https://www.implicator.ai/claudes-rate-limits-arent-a-capacity-problem-theyre-a-math-problem/)