The monthly fee is only the first number
This guide uses official rules checked on September 30, 2026, not a private subscriber's test log. The Pro page lists $20 per month in the US. The Max page lists $100 for Max 5x and $200 for Max 20x on the web, with Max currently monthly-only. Local currency, taxes, and mobile checkout may differ.
| Plan | Monthly price | Session allowance |
|---|---|---|
| Pro | $20 | Reference allowance |
| Max 5x | $100 | 5 times Pro |
| Max 20x | $200 | 20 times Pro |
The Max multipliers refer to per-session usage allowance, not a guaranteed multiplication of completed projects. Pro includes Claude Code access but not Console API usage, which is billed separately.
Two reset clocks can stop the same workflow
Pro and Max have five-hour session resets and weekly limits across models. Their help pages say the weekly reset is an assigned fixed day and time, independent of when you began using Claude or started the subscription. Check Settings > Usage for your account's next reset. Additional model or feature caps may apply.
A session reset does not imply the weekly allowance has also reset. If work is still blocked, read which allowance is exhausted before assuming the payment failed. Keep the two clocks separately in your work log.
Claude and Claude Code share consumption
The usage-limit explanation says claude.ai, Claude Code, and Claude Desktop count toward the same limit. Model, effort, conversation complexity, and features affect consumption. A context-length limit is a different constraint from the allowance available across a time period.
In practice, a measurement made while another agent is working in parallel cannot cleanly assign the observed change to one task. Pause unrelated work for the comparison, or label the result as combined consumption. Starting a fresh conversation is not a new subscription allowance.
A model in the picker may use a different billing path
Fable's plan rules are an important exception. On Pro, Fable 5 and 5.1 use pay-as-you-go credits rather than the included allowance. On Max, Fable can use up to 50% of the regular weekly limit; this is part of that limit, not 50% extra. Continuing through paid credits adds spending.
Check that distinction before a long run. The presence of a model name alone is not proof that the next task is covered by the monthly fee. A budget-conscious workflow should record both remaining included capacity and any paid-credit consumption.
How to build an actual usage log
- Record the plan, chosen model, effort, session and weekly readings, and both displayed reset times before work starts.
- Define one bounded outcome, such as an article passing source review or a fix passing its focused tests.
- Record the readings after completion, plus retries, paid credits, and human checking time. Include failed attempts.
- Repeat for routine and difficult tasks separately, then report the range of consumption per accepted result.
Illustration only: a weekly reading that moves from 40% used to 46% used is a six-percentage-point observation for that work block. It is not six dollars or a portable task price. Rounded displays, concurrent work, and resets can make a single observation unsuitable for extrapolation.
Choose the plan from interruptions, not multipliers alone
Track which limit actually halted useful work. A larger session allowance can address a session bottleneck without proving that the weekly bottleneck disappears. Also record how much unused capacity remains at reset, so peak-day frustration is not mistaken for consistently high demand.
Use the Claude API pricing article for metered API budgets. The ChatGPT subscription comparison uses its own current rules; do not import a five-hour assumption from one provider into the other.