The quick verdict
Stay on Pro ($20 a month, or $17 billed annually) if Claude Code is a side tool and you rarely hit limits. Move to Max ($100) if Claude Code is your main development tool and you hit Pro's 5-hour or weekly limit most weeks. Move to Ultra ($200, formerly Max 20x) if you run Claude all day, lean on Opus, or run several agents in parallel: verified Ultra accounts have no 5-hour limit at all, only weekly limits. Use the API instead if most of your tokens come from automated background jobs rather than interactive coding.
What each tier actually gives you
On 30 October 2026 Anthropic renamed Max 5x to Max and Max 20x to Ultra. Pro and Max cover the Claude apps and Claude Code under a 5-hour limit and a weekly limit, and Max gives 5x the usage of Pro in both. Ultra gives 10x Pro's weekly usage and, once Anthropic has verified the account, no 5-hour limit at all; until then a temporary session limit applies. All plans draw from one shared allowance across web, desktop, mobile, and Claude Code, and paid plans can turn on usage credits to keep working at standard API rates after a limit. Max also includes $100 a month and Ultra $200 a month of Claude API credits. They cover the Claude API, Managed Agents and the Agent SDK, not Claude Code, and they do not roll over.
The break-even at October 2026 prices
A typical Claude Code turn sends about 16,000 input tokens and gets about 1,500 back. Without caching that costs about $0.047 on Sonnet 5.5 ($2 and $10 per million) and $0.094 on Opus 5.5 ($4 and $20). So $100 buys roughly 2,100 uncached Sonnet turns or 1,060 Opus turns. Claude Code caches most of the repeated context, which cuts a turn to roughly $0.02 on Sonnet 5.5 and $0.04 on Opus 5.5, so with caching the break-even rises to about 5,000 Sonnet turns or 2,500 Opus turns a month. Spread over 21 working days, Max pays off somewhere between 100 and 250 turns a day, depending on model and how well your sessions cache.
Why the answer changed this fall
Sonnet 5.5 costs $2 and $10 per million tokens and Opus 5.5 $4 and $20, down from $3 and $15 for Sonnet 4.6 and $5 and $25 for Opus 4.8. Cheaper tokens mean you need more usage before a flat subscription beats pay-as-you-go, so some moderate users who were right to buy Max in the summer are now better off on Pro plus occasional usage credits. Heavy users still win big on Max or Ultra, and the monthly API credits add value if you would have paid for API access anyway.
When the API is the cheaper choice
Max plans are built for interactive use, not high-volume automation. If your tokens come from background pipelines, batch jobs, or a product calling the API around the clock, pay-as-you-go with prompt caching and the 50 percent Batch API discount is often cheaper and does not bump into interactive limits. If you barely use Claude, Pro or even the free tier is plenty.
The factor most comparisons miss
The real question is rarely Max versus the API in isolation. It is how many AI plans you pay for at once. Plenty of builders carry Claude Max, ChatGPT Plus, and Cursor at the same time, well over $130 a month combined, and most of it stays invisible until the cards are charged. Before upgrading one plan, see the whole stack in one number.
Check your own break-even instead of guessing
Tokens 4 Breakfast reads your real Claude Code usage and shows what it would have cost at API rates, next to your plan price and every other AI subscription, in the macOS menu bar. Pro adds Plan Value, which weighs your actual usage against what you pay so the Max-versus-Pro decision rests on your numbers. Free for one provider, one-time $14.99 for the full picture.
Frequently asked questions
Is Claude Max worth it?
If Claude Code is your main tool and you run it most of the working day, yes. At current prices, Max ($100) pays for itself at roughly 2,000 to 5,000 Claude Code turns a month on Sonnet 5.5, or about half that on Opus 5.5. If you use Claude Code a few times a week, Pro at $20 is the better deal.
Claude Max or Ultra: which is worth it?
Max ($100, formerly Max 5x) suits developers who hit Pro's limits most weeks. Ultra ($200, formerly Max 20x) suits people who use Claude all day, rely on Opus, or run several agents in parallel, and verified Ultra accounts have no 5-hour limit, only weekly limits. If you hit Max limits regularly, Ultra usually costs less than paying API rates for the overflow.
Does Claude Max include API credits?
Yes. Max includes $100 a month and Ultra $200 a month of Claude API credits, and Team plans share up to $500. They cover the Claude API, Managed Agents and the Agent SDK, but not Claude Code or extra usage, and unused credits do not roll over. If you would pay for API access anyway, they make the plan worth more than its usage limits alone.
How much usage do Claude Pro, Max, and Ultra give you?
Anthropic sells usage as multiples of Pro rather than a fixed token count. Max gives 5x Pro's usage, with a session limit that resets every five hours and a weekly limit. Ultra gives 10x Pro's weekly usage and, on verified accounts, no 5-hour limit. All plans share one allowance across the Claude apps and Claude Code, so real headroom depends on your model mix and session length.
When is the API cheaper than a Claude Max subscription?
When most of your tokens come from automated background jobs rather than interactive coding, or when you use Claude lightly. Pay-as-you-go with prompt caching and the 50 percent Batch API discount is often cheaper there, and it avoids interactive rate limits.