Claude Max is $100/mo for five times Pro session capacity or $200/mo for twenty times, as of August 2026. The models and features are identical to Pro; only the allowances differ. Max 5x suits one engineer working with an agent most of the day. Max 20x is for parallel sessions, long autonomous runs, agent teams, or defaulting to Opus. Between the two sits a third option most people miss: stay on 5x and turn on usage credits for the heavy weeks.
- Max 5x is $100/mo, Max 20x is $200/mo. The multiplier is against Pro, not against each other.
- Anthropic describes the multiplier as usage per session. Weekly caps are separate numbers and do not scale the same way.
- Max buys headroom only. Same models, same Claude Code, same features as Pro.
- The honest test for 20x is parallelism: if you run one session at a time, 5x is usually enough.
- Max 20x at $200 is still cheaper than the API for anyone doing all-day agent work.
- Usage credits on 5x are the middle rung: metered overage at API rates with a cap you set.
What the two tiers are
Claude Max tiers, read from Anthropic in August 2026. Multipliers are against Pro.
| Tier | Price | Session capacity | Typical fit |
|---|---|---|---|
| Pro | $20/mo | Baseline | Part-time agent use, mostly chat. |
| Max 5x | $100/mo | 5x Pro per session | One engineer, agent-assisted most of the working day. |
| Max 20x | $200/mo | 20x Pro per session | Parallel sessions, long autonomous runs, agent teams, Opus by default. |
| Team premium seat | $100/seat/mo annually | 5x standard seat | The same money as Max 5x, with admin controls attached. |
The three limits still apply
Both Max tiers have the same three-limit structure as Pro: a rolling five-hour session allowance, a weekly cap across all models, and a separate weekly cap on Opus. Claude Code names which one you hit. If the message says Opus, run /model and switch; you keep working immediately. If it says session or weekly, the limit is shared across models and switching does nothing.
The honest test for which one
Five questions. Two or more yes answers points at 20x.
- Do you run more than one agent session at a time? Parallelism multiplies consumption directly and is the clearest signal.
- Do you leave agents running on long autonomous tasks? A 40-minute unattended run consumes far more than 40 minutes of supervised back-and-forth.
- Do you use agent teams? Anthropic puts them at roughly 7x the tokens of a standard session when teammates run in plan mode, because each teammate carries its own context window.
- Do you default to Opus rather than escalating to it? Opus burns allowance several times faster than Sonnet, and it has its own weekly cap on top.
- Do you work in a large repo where the agent reads a lot before it acts? Context size drives consumption more than session length does.
If all five are no, 5x will very likely carry you, and you have saved $1,200 a year. If three or more are yes, 20x is not indulgence; you will meet the 5x ceiling regularly and lose more in interruption than the difference costs.
The rung between 5x and 20x
Most people treat this as a binary. It is not, since usage credits shipped. On Max 5x you can turn on credits, set a monthly spending cap, and let anything past the included allowance bill at standard API rates. That converts an occasional two-week crunch from a $100 permanent upgrade into a metered top-up you can watch.
The arithmetic: if your credit spend regularly exceeds about $100 a month, 20x is cheaper and simpler. Below that, 5x plus credits wins, and it wins by more if your heavy weeks are seasonal rather than constant.
Max against the API
Max 20x at $200 sounds expensive next to a metered key until you price the same work.
A representative heavy month, priced both ways at Sonnet 5 standard rates.
| Usage | On Max 20x | At API rates |
|---|---|---|
| 6 hours/day of Sonnet agent work, 22 days | $200 | roughly $400 to $900 |
| Same, Opus heavy | $200 | roughly $900 to $2,500 |
| Two parallel sessions, all day | $200 | roughly $1,500 to $4,000 |
| Anthropic's published enterprise average | n/a | $150 to $250 per developer per month |
That last row is worth sitting with. Anthropic's own documentation puts the average across enterprise deployments at about $13 per developer per active day and $150 to $250 per developer per month on metered billing, with 90 percent of users under $30 per active day. The average metered developer is already spending more than Max 5x costs.
The gap is not marginal. For interactive daily work the subscription is dramatically cheaper, which is the intended design: Anthropic prices the subscription for humans at keyboards and the API for programs.
The API wins on shape rather than volume. If your work is scripted, scheduled, bursty, or needs to run without a person present, a key is the right tool regardless of the arithmetic.
What Max does not fix
- Windows still exist. 20x is twenty times Pro per session, not unlimited, and the weekly caps do not scale by the same factor. Sustained parallel Opus work can still meet one.
- The Opus cap is separate. Upgrading raises it; it does not remove it. Defaulting to Opus will still find it.
- It is one account. Two people cannot share it, and the windows are per account regardless.
- It does not make the model better. If output quality is the problem, more headroom will not help.
- It does not tell you where you are. The plan raises the ceiling;
/usageonly reports the machine you are typing on, so a two-laptop setup reads low on both.
Questions people ask
$100 a month for Max 5x, or $200 a month for Max 20x, as of August 2026. Anthropic describes the multipliers as five times and twenty times more usage per session than Claude Pro.
Session capacity only. Identical models, identical features, identical Claude Code. 20x gives you four times the per-session allowance of 5x. The weekly caps above the session window are separate numbers and do not scale by the same factor.
If you run parallel sessions, use agent teams, leave agents on long autonomous tasks, or default to Opus, yes: the same work on an API key typically costs several times more. If you run one supervised Sonnet session at a time, 5x is usually enough, and 5x plus usage credits covers the occasional heavy week for less.
No. It raises them. The rolling five-hour session window, the weekly cap, and the separate Opus cap all still apply, at 5x or 20x Pro session allowance.
Yes, they are monthly subscriptions. Running 5x for a month and counting your interruptions with /usage is the cheapest way to find out whether you need 20x.
No. Subscriptions and the API are separate billing systems and a Max subscription does not fund an API key. The one bridge is usage credits, which bill subscription overage at standard API rates.
Compare your monthly credit spend against $100. Above roughly $100 of credits a month, 20x is cheaper and involves no thinking. Below it, especially if your heavy weeks are seasonal, 5x plus a capped credit budget costs less.
Yes, and that matters for the arithmetic: the allowance is shared across Claude chat, Claude Code, and Cowork. Heavy chat use draws from the same window your agent sessions do.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.