Claude Pro vs Max: we measured how much one coding job uses
Claude Pro costs $20, Max $100 or $200. We measured how much of a Max 5x plan one small coding job uses, and what that means for each plan.
Claude Pro costs $20 a month. Claude Max costs $100 or $200. Both give you the same Claude models and Claude Code. The big difference is how much you can use them before you hit a limit. Anthropic does not say how many messages or jobs that is. So we measured it: on our own Max 5x plan, one small coding job with Sonnet 5.5 used about 0.22% of the 5-hour limit. That works out at about 90 jobs per 5 hours on Pro, about 450 on Max 5x and about 1,800 on Max 20x.
Pro vs Max: what is different?
Anthropic's pricing page and help page list the plans. These are the US prices before tax on 12 October 2026:
| Pro | Max 5x | Max 20x | |
|---|---|---|---|
| Price | $20 a month, or $200 a year | $100 a month | $200 a month |
| Use per 5 hours | the base | 5 times Pro | 20 times Pro |
| Claude Code | yes | yes | yes |
| Fable 5.1 | only with paid extra credits | up to half your weekly limit | up to half your weekly limit |
| Monthly API credit | no | $100 | $200 |
| Priority when busy | no | yes | yes |
Max is billed monthly only. Pro and Max use the same models, so Claude is not smarter on Max. You just get more of it. The API credit is for building your own apps through the Claude API, the paid connection that programs use to talk to Claude. It does not pay for Claude Code or the Claude apps, Anthropic says.
How the limits work
Every paid plan has two limits. The first is per session of 5 hours: when you hit it, you wait until the session resets. The second is per week, and it counts all models together. Everything you do in the Claude app, on the website and in Claude Code comes out of the same pot.
Anthropic only says how the plans compare: Max 5x gives five times Pro's use per session, Max 20x twenty times. It does not give a number of messages, because that depends on how long your chats are, which model you use and what you ask. You can see where you stand under Settings and then Usage, or with /status in Claude Code.
When you hit a limit, you can wait, move to a bigger plan, or turn on usage credits. Those let you keep going, and you pay for that extra use at API prices.
Our test: how much one coding job uses
While Claude Code works, it quietly reports how much of your limit is used, as a number between 0 and 1. We used that to measure. On 12 October 2026 we gave Claude Code our three small coding jobs from our secret test set, three times each. We did that with three models: Sonnet 5.5, Haiku 4.5 and Opus 5.5. That is 27 jobs on the site owner's Max 5x plan, through the Agent SDK. Nobody used the plan for anything else during the test.
| Model | Secret checks | 5-hour limit, 9 jobs | Per job |
|---|---|---|---|
| Sonnet 5.5 | 186 of 186 | 19% → 21% | about 0.22% |
| Opus 5.5 | 186 of 186 | 31% → 35% | about 0.44% |
| Haiku 4.5 | 158 of 186 | 21% → 31% | about 1.11% |
All 27 jobs together moved the weekly limit from 73% to 75%. Claude Code only reports whole percents, so each number can be one point off. That is why we give ranges below.
What that means for Pro, Max 5x and Max 20x
If one Sonnet 5.5 job uses about 0.22% of a Max 5x session, a full session fits about 450 jobs like ours. Pro has a fifth of that and Max 20x four times as much, by Anthropic's own numbers:
| Model | Pro | Max 5x | Max 20x |
|---|---|---|---|
| Sonnet 5.5 | about 90 | about 450 | about 1,800 |
| Opus 5.5 | about 45 | about 225 | about 900 |
| Haiku 4.5 | about 18 | about 90 | about 360 |
These are jobs per 5-hour session. For Sonnet the range is wide: 60 to 180 on Pro. And our jobs were small, about half a minute each. A real coding session with a big project and a long chat uses much more per step, so expect fewer.
The weekly limit matters too. On Max 5x, our 27 jobs used about 2% of the week, so roughly 1,400 small jobs a week (between 900 and 2,700). Anthropic does not say how the weekly limits of Pro and Max compare, so we do not guess that for Pro.
A cheaper model is not always lighter on your plan
We expected the small Haiku 4.5 to use the least of the limit. It used the most: about 1.11% per job, five times as much as Sonnet 5.5. The reason was the work, not the price. Haiku needed many more steps to finish: about 441,000 tokens per job against 171,000 for Sonnet. Tokens are the small pieces of text an AI reads and writes. And Haiku still missed 28 of our 186 secret checks.
Opus 5.5 is the bigger model. It used about as many tokens as Sonnet, but about twice as much of the limit per job. It took a bit longer (53 seconds against 34) and passed every check.
So on a plan, the model that finishes a job in the fewest steps is often the cheapest. In our test that was Sonnet 5.5.
Fable 5.1: included on Max, extra on Pro
Fable 5.1 is Anthropic's biggest model. This is where Pro and Max differ most, apart from the amount of use.
On Max, Fable is part of the plan: you can spend up to half of your weekly limit on it at no extra cost, says Anthropic. It uses up the limit faster than other models. On Pro, Fable is not part of the plan at all. You can only use it with usage credits, so you pay extra for every bit.
If you want to work with Fable often, Max is the only plan where that does not cost extra. We did not test Fable on the plan here.
Is Claude Max worth it?
It depends on how often you hit the limit:
- You code a few times a week: Pro. By our estimate it fits about 90 small jobs per 5 hours with Sonnet 5.5, which is plenty for short sessions.
- You hit Pro's 5-hour limit most days: Max 5x. Five times the room for five times the price.
- You run long agent sessions, or several at once: Max 20x.
- You want Fable 5.1 a lot: Max, because only there it is part of the plan.
- Your busy days are rare: stay on Pro and turn on usage credits when you need them. At API prices, one of our small Sonnet jobs was worth about 12 cents.
The $100 or $200 API credit on Max is a bonus if you also build your own apps. It does not give you more room in Claude Code. Our Claude Code pricing article explains the API prices behind all this, and our AI API price comparison compares them with other providers.
Questions people ask
What is better, Claude Pro or Claude Max?
Neither gives you a smarter Claude: both have the same models. Max gives you 5 or 20 times more use per 5-hour session, Fable 5.1 inside the plan, and a monthly API credit. In our test, Pro would fit about 90 small coding jobs with Sonnet 5.5 per 5 hours, Max 5x about 450.
What does Claude Max get you?
Everything in Pro, plus 5x or 20x more use per 5-hour session, higher output limits and priority when it is busy. You can also use Fable 5.1 for up to half of your weekly limit at no extra cost. And you get $100 or $200 a month in API credit, but that does not pay for Claude Code.
Is Claude Code Max 20x worth it?
Only if you keep hitting the Max 5x limit. By our estimate Max 20x fits about 1,800 small Sonnet 5.5 coding jobs per 5 hours. Most people run into other limits first: the weekly limit also applies, and real coding sessions are bigger than our test jobs.
Is it worth buying extra usage on Claude?
For the odd busy day, yes. Usage credits are charged at API prices after you hit a limit. One of our small coding jobs with Sonnet 5.5 was worth about 12 cents at those prices. If you need extra almost every day, moving to Max is usually cheaper.
How to get Claude Max for cheap?
Anthropic says it has no standing discounts on Max, and its support team cannot give coupons. Promotions are sometimes announced through its own channels. The only fixed saving is on Pro: about $17 a month if you pay for a year.
Is $100 Claude worth it?
Max 5x at $100 is worth it if you often hit the 5-hour limit on Pro. In our test it fits about 450 small Sonnet 5.5 coding jobs per 5 hours, five times as many as Pro. It also includes Fable 5.1 and a $100 monthly API credit for your own apps.
How we checked
We read Anthropic's pricing page and its help pages on the Max plan, Claude Code with Pro or Max and Fable on your plan on 12 October 2026. The prices are US prices before tax. The screenshots are our own and only cropped.
For the measurement we used the site owner's Max 5x plan on 12 October 2026, between 16:12 and 16:41 UTC. We ran 27 jobs: three coding jobs, three tries each, with Sonnet 5.5, Haiku 4.5 and Opus 5.5, effort "medium", through Claude Code's Agent SDK with a clean setup. During every step, Claude Code sent a message with the share of the 5-hour and weekly limit already used (see the image). We compared the first and last value of each batch. The owner did not use Claude during the test, and our own work paused.
This gives estimates, not promises. Claude Code reports whole percents. Our jobs were small. And Anthropic can change the limits at any time, as its help page says. The charts were made by us. None of the images was made by AI.