Claude Code, Codex and Cursor all top out at $200 a month. The bills don't.

Same ceiling, very different bills. The headline price hides the real difference, which is how each one meters you.

The options

Cursor now splits usage into two pools: its own first-party models, and third-party models billed at their underlying API rates. Choosing a frontier model burns your allowance faster. The catch: the more you reach for the best model, the sooner the month's allowance runs out.

GitHub Copilot moved to usage-based credits on 1 June, and Opus-tier models consume three times the premium-request allowance. Same shape of catch: the model you pick decides how long your credits last.

Claude Code's $100 and $200 tiers put a firmer ceiling on spend and trade overage risk for rate limits. The catch: on a heavy day you can hit a limit and wait, rather than pay more and carry on.

OpenAI sells Codex through ChatGPT's plans, from the $20 entry plan up to the $200 Pro tier, with a $100 option in between. If your team already pays for ChatGPT, check what your plan includes before buying anything new.

What it costs on your worst week

Ask each tool one question: what happens on your worst week? The big refactor, the framework migration, the test-fix loop that runs all afternoon.

A flat plan that throttles you is predictable: you know the bill, and you lose some hours. A metered plan that keeps going is productive right up until the invoice arrives. Pick the failure you can live with.

Some sums for a small team (FER calculations from the list prices above):

• Three developers on $20 entry plans: $60 a month, or $720 a year.
• Three developers on $200 top tiers: $600 a month, or $7,200 a year.
• The difference is $540 a month. Spread across three people, the bigger plan needs to save each of them a few hours a month to earn its keep.

The point isn't that one number is right. It's that "should we upgrade?" is a sum you can do, not a feeling you have on a bad afternoon.

What we'd pick

• **Solo developer, or a founder who codes:** start on a $20 plan for a month and note every time you hit a limit or wished for a better model. Upgrade on that log, not on a mood.
• **Two to five developers:** let one person trial a top tier for a month while the others stay on entry plans. Compare what got shipped, not how it felt.
• **Agency or dev shop billing clients:** prefer the predictable plan. A surprise overage you can't pass on comes straight out of your margin, and a rate limit is easier to plan around than an invoice.
• **Everyone:** keep prompts, rules and project context in the repo rather than in one tool's settings, so switching costs you a day rather than a month.

How to run the one-month trial

Keep it simple enough that you'll actually do it:

1. Before you start, write down the question. "Does the $200 tier save Sam enough time to justify $180 a month more?" is a question. "Is it better?" isn't.
2. Each Friday, the person on trial notes three things: how often they hit a limit, which tasks went noticeably faster, and anything they shipped that they wouldn't have tried before.
3. At the end of the month, put the hours saved next to the extra cost. If you bill by the hour or by the project, use the rate you'd charge a client.
4. Decide, and set a reminder to check again in six months. Prices in this market move faster than annual plans.

What we'd skip: paying for two top tiers indefinitely "to compare". Run the month-long trial with a question you want answered, then pick one.

Disclosure: this publication's research agent runs on Claude.