Almost every serious AI coding tool now lands at the same Pro price: $20 a month. Cursor, Windsurf, and Claude Code all sit there, and GitHub Copilot Pro is even cheaper at $10. The convergence makes it look like a commodity. It is not. The same $20 buys four genuinely different things, because the four tools meter you in four different ways, and the cheapest tool for you is whichever meter happens to match how you actually work.
This is a guide to the meters, not the models. Benchmark scores change every few weeks and there are a hundred articles tracking them. The billing mechanic underneath each tool changes far less often and decides your bill far more, and almost nobody writes it down. So here are the four mechanics in plain terms, what runs out under each, and how to tell which one fits your pattern.
The four meters
1. Credit pools (Cursor)
Cursor Pro at $20 a month gives you a credit pool worth about $20 of model usage, plus an "Auto" mode that routes to a managed model and stays unlimited on every paid tier. Heavier tiers buy bigger pools: Pro+ at $60 a month (a 3x pool) and Ultra at $200. The mental model is a prepaid balance. You spend it on whatever models and modes you choose, and when it runs dry you either drop to the unlimited Auto model or top up. The flexibility is the selling point: you decide where the budget goes. The risk is that a few heavy frontier-model sessions can drain a month's pool in days if you are not watching.
2. Daily and weekly quotas (Windsurf)
Windsurf overhauled its pricing on March 19, 2026, dropping its old credit system for daily and weekly quotas, and raised Pro from $15 to $20 to match the field. A new Max tier arrived at $200. A quota is not a balance you spend, it is a ceiling you hit. You get a certain amount of activity per day and per week, and when you reach it you wait for the window to reset. For steady, all-day usage this is predictable and calm. For someone who works in intense multi-hour bursts, hitting a daily cap at 2pm and being locked out until tomorrow is the failure mode to watch.
3. Time-windows (Claude Code)
Claude Code meters by rolling time windows rather than credits or daily quotas. Pro at $20 gives roughly 44,000 tokens per 5-hour window; Max 5x at $100 gives about 88,000 per window; Max 20x at $200 gives around 220,000. On top of the 5-hour window sits a weekly cap on active compute, and that weekly bucket is shared across Claude Code, the Claude chat app, and Cowork. Anthropic has been actively loosening these: it doubled the 5-hour limits on May 6, then added a 50 percent weekly-limit increase on May 13 that runs through July 13, 2026. The shape rewards a rhythm of focused sessions with breaks. It punishes someone who needs one continuous eight-hour grind, because the window, not your wallet, is the constraint.
4. Per-token credits (GitHub Copilot, from June 1)
Copilot is the newest convert. Through May it ran on premium requests (a fixed monthly count of non-completion actions, with overages around $0.04 each). On June 1, 2026 it switches to GitHub AI Credits: 1 credit equals $0.01, and each request burns credits at the underlying model's per-token rate. Code completions stay free. This is the most directly usage-based of the four, and also the most exposed to model choice, because the per-model multipliers swing hard (Opus 4.7's reportedly jumps to 27x). We walk the credit math in the Copilot billing breakdown.
The same $20, side by side
| Tool | Pro price | Meter | What runs out | Best fit |
|---|---|---|---|---|
| Cursor | $20/mo | Credit pool (+ unlimited Auto) | Your prepaid balance | Mixed usage where you want to choose where budget goes |
| Windsurf | $20/mo | Daily + weekly quotas | Activity allowance per window | Steady, spread-out daily work |
| Claude Code | $20/mo | Rolling time-windows + weekly cap | Tokens per 5-hour window | Focused sessions with natural breaks |
| GitHub Copilot | $10/mo | Per-token AI Credits (June 1) | Credit balance, by model rate | Completion-heavy work; light agent use |
Match the meter to how you work
Forget which model benchmarks highest this month. Ask which of these sentences sounds like your day:
- "I type code and accept suggestions most of the day." Completions are free almost everywhere, so the cheapest credible option wins. Copilot Pro at $10 is hard to beat, and the June 1 change barely touches you.
- "I run agents in long bursts, then nothing for hours." Avoid daily quotas (you will hit the wall mid-burst) and avoid tight time-windows. A credit pool you can spend on your own schedule fits best.
- "I work steadily all day, every day." Quotas reward you, because spread-out usage rarely hits a daily ceiling and the price is fixed regardless of how the balance is spent.
- "I do deep multi-hour refactors on frontier models." This is the expensive corner. Time-windows throttle you and per-token credits bill you the full frontier rate with the heavy multipliers. Budget for a Max-style $100 to $200 tier, or split the work so caching can fire.
The part the tools do not solve for you
Every one of these meters answers the vendor's question (how do we limit one seat) and ignores yours (what is this costing across the team, by person, by project). A credit pool, a daily quota, and a 5-hour window are three different units, so the moment you run more than one tool, or more than a handful of engineers, you cannot compare or roll them up. That is a usage-metering problem, and it is the same one the tool vendors are wrestling with on their side of the meter. If you want one normalized view of spend across tools and people, with caps and showback, that has to live above the individual tools. We cover the engineer-level version in the per-engineer cost-cap playbook, and the cautionary version in the Claude Code budget-bomb case.
Common questions
Which is actually cheapest? For completion-heavy work, Copilot Pro at $10. For everything agentic, there is no universal answer, because the cheapest tool is the one whose meter never makes you hit a wall mid-task. Measure your own week of usage before committing.
What happens when I hit the limit? Credit pools let you top up or fall back to a free model. Quotas lock you out until the window resets. Time-windows throttle until the next 5-hour or weekly reset. Per-token credits let you keep going and keep billing. Only one of those four can produce a surprise invoice, and it is the last one.
Do these prices include the model API cost? Yes, the subscription bundles model access. That is exactly why the meter matters: the vendor is absorbing variable token cost behind a fixed or semi-fixed price, and the meter is how they keep that bet from blowing up. When the meter tightens, that is the vendor's variable cost showing through.