Guides · September 2026
Claude plan calculator: Pro vs Max vs API
Put in how much your agents actually run and get back the tokens, the list-price cost, and which plan plausibly covers it — with every assumption on the table.
Three plans, one pool of usage.
Claude Pro is $20 a month ($17 on annual billing). Max comes in two sizes: Max 5x at $100 a month and Max 20x at $200. Anthropic defines the multiplier per session: Max 5x gives five times more usage per session than Pro, Max 20x twenty times. That is the whole published quota — a ratio, not a token count.
A session is a five-hour window. On top of the session limit, every paid plan has a weekly limit that applies across all models, resetting at a fixed time each week assigned to your account, and Claude Code reports separate weekly caps per model family. Everything you do in the Claude apps and in Claude Code draws on the same limits.
Claude Code is included on all three. What differs is the default model: on Pro it starts on Sonnet 5, on Max on Opus 5, and /model switches either way. At API list prices Opus 5 costs two and a half times Sonnet 5 per token — $5 in and $25 out per million against $2 and $10 — which is the main reason Max subscribers burn through a session faster than the multiplier led them to expect.
When the pool runs dry you can wait for the reset, or turn on usage credits: prepaid funds billed at standard API rates, under a monthly spend limit you set. Bundles of $50, $250, and $1,000 take 10, 20, and 30 percent off those rates.
The calculator.
Nothing is stored and nothing is sent; the sums run in your browser. Every number that is not Anthropic’s is editable under “Assumptions”.
Hours an agent is actually working — not hours the terminal is open.
Pro starts Claude Code on Sonnet 5, Max on Opus 5. Opus is 2.5× the price per token.
Informational: the pool is per account, so every machine you log in on draws from the same limits. Count the hours across all of them above.
Active hours a month
87
Tokens a month
750M
693M cached · 52M fresh · 4.3M out
API cost at list, a month
$501
$5.78 per active hour at this mix
Pro
$20 a month
Won’t cover itPlausibly covers $40–$150 of API-equivalent use a month.
Past the band: $351–$461 a month more as usage credits at API rates.
Max 5x
$100 a month
Probably covers itPlausibly covers $200–$750 of API-equivalent use a month (Pro’s band × 5).
Past the band: $0–$301 a month more as usage credits at API rates.
Max 20x
$200 a month
Covers itPlausibly covers $800–$3,000 of API-equivalent use a month (Pro’s band × 20).
Max 5x at $100 would probably cover this, but the overflow it risks — up to $301 a month in usage credits — costs more than the $100 step up to Max 20x, which buys 4× the pool. Max 20x at $200 is the better buy, and it covers this comfortably.
The bands are our anchor for Pro scaled by Anthropic’s published multipliers, not published quotas. Limits are enforced per five-hour session and per week, so bursts hit them before a monthly average says they should.
›Assumptions — edit them
Tokens per active hour, by kind. Claude Code re-sends the conversation on every request and an agentic turn makes many requests, so cache reads dominate. Edit any field and the intensity becomes “Custom”.
What Pro’s included usage is worth per month at API list prices. Anthropic publishes the prices and the 5× and 20× multipliers, not token quotas; this anchor is our estimate, and the Max bands are it times the multiplier.
- Published, used as is
- Pro $20, Max 5x $100, Max 20x $200 a month. Max 5x is 5× Pro’s usage per session, Max 20x is 20×. Sonnet 5: $2 in, $0.2 cache read, $10 out per million tokens. Opus 5: $5, $0.5, $25. A month is 52⁄12 weeks.
- Assumed, editable above
- Tokens per active hour for light, typical, and heavy work; that plan usage scales with the API-list value of what you send, so an Opus hour draws 2.5× a Sonnet hour; and the Pro anchor of $40–$150.
How Claude Code spends a plan.
Every turn sends the conversation so far, the project context, and the new prompt, and an agentic turn is many requests — one per tool call round. That is why the calculator counts context traffic per active hour rather than what you typed: a one-line question in a session that has been open all day still carries the whole day. Prompt caching makes the re-sent part cheap — a cache read is a tenth of the input price — and on a subscription the cache lives an hour, dropping to five minutes once you are drawing on usage credits.
The cheapest habits follow from that. /clear between unrelated tasks, so the next request does not carry the last one. Sonnet for the routine work and Opus where the reasoning earns its price. Fewer MCP servers in context. And the check itself, from inside the CLI:
> /usage
> /model
> /usage-credits
# plan bars, attribution, and a 24h / 7d breakdown; the model picker; credits on claude.ai
The errors are literal. You’ve hit your session limit is the five-hour window; You’ve hit your weekly limit is the account-wide cap; You’ve hit your Opus limit is a per-model weekly cap, and switching to Sonnet with /model keeps you working.
Is extra usage worth it, or the next plan up?
The arithmetic is short. Max 20x gives twenty times Pro’s usage for ten times the price — twice the usage per dollar of Pro or Max 5x. So for anyone at the top of Max 5x the answer is nearly always the upgrade: $100 more buys four times the pool, while $100 of credits buys $100 of tokens at list, or up to about $143 worth with the 30-percent bundle. The calculator does this comparison for you, with the overflow it estimates against the step up.
Credits earn their place in three cases.
You are already on Max 20x
There is no next plan. The weekly cap is the ceiling, and credits at API rates — cheaper by the bundle — are how the rest of the month gets paid for.
The spikes are rare
A release week that runs three agents overnight is not a reason to pay $100 more every month. Turn credits on with a spend limit that matches the spike.
You want the bill to be legible
Credits show up as dollars in /usage and on the usage page; the plan’s own pool does not. Some people want the meter even when the plan would have covered it.
One thing the calculator cannot see is concurrency. Three sessions on three machines draw down the same five-hour window at three times the rate, so the weekly cap arrives on a Wednesday. If that is your shape, count all of the hours above, and weight the answer toward the bigger plan.
Whichever plan you pick, the CLI logs in on the host.
The subscription is attached to your account, and Claude Code authenticates on whatever machine it runs — the desk box, the GPU rig, the VPS — each one a claude login away from the same pool. Nothing about that changes with spawnd. Agents authenticate on the host, as always; spawnd holds no provider credentials and adds no API-key markup, so a Pro or Max plan is used exactly as Anthropic’s CLI uses it.
What changes is where you can be. One daemon per host you own dials out, so nothing listens on the host and no VPN is needed; any browser is the console, and on a phone it installs to the home screen as a web app. The sessions live on the host — a worker process owns each PTY — so the run you started at the desk keeps going after the tab closes, and the permission prompt an agent raises at nine reaches your phone. Every machine you own shares the one subscription, from any device you have approved. Orchestrating several agents at once is where that starts to matter; the Claude Code pillar has the rest.
Real PTYs owned by host workers. Hosts dial out — zero open ports. Your browser talks to each daemon peer-to-peer, end-to-end encrypted. Open source, MIT / Apache-2.0. Read the threat model.
Start
One subscription, every host you own.
curl -fsSL https://spawnd.dev/install.sh | shQuestions
- Does Claude Code on Pro use Opus?
- Pro starts Claude Code on Sonnet 5; Max starts it on Opus 5. Run /model to see what your account offers and switch. Opus costs two and a half times as much per token at API rates, and it draws a session down at about that rate too.
- Do the limits reset monthly?
- No. There is a session limit that resets every five hours and a weekly limit that resets at a fixed time each week, plus per-model weekly caps. The calculator averages to a month so the numbers can sit next to the prices; a bursty week reaches the cap sooner than the average suggests.
- Is the API cheaper than a Max plan?
- For light or occasional use, often yes — the calculator prints the list-price cost next to every plan and flags when it undercuts the plan you would need. Past a couple of hours a day the plans win, and Max 20x is the best value per dollar of the three.