The short version: most small business owners do fine on the $100 per month plan for the whole month. Heavy operators running multiple agents and big projects move to $200. Usage is metered in two windows at once, five-hour and weekly, and both refill on their own. The only way to get hurt is buying extra credits mid-cycle, because those are billed far above your subscription's subsidized rate.
I walk through the actual usage and billing screens in the video, starting with that client text:
The Two Meters Running at Once
Click your name in the bottom left of the app and open Usage. You will see your consumption measured two ways:
The five-hour window. A rolling short-term allowance. A heavy building session can exhaust it, and then it refills a few hours later. My client ran hers dry in a morning of building; when she logged back in that evening, it had reset because her weekly allowance still had room. That is the meter most "out of credits" panics are actually about.
The weekly window. The bigger bucket behind it. Burn through this and the five-hour refills stop mattering until the week turns over.
One wrinkle worth knowing: five hours of allowance late on a Friday night stretches further than five hours midday Monday, because the meter responds to how loaded the system is when you use it. Schedule your heavy builds off-peak and the same plan goes further.
Resets: The Free Refills Most Owners Never Notice
OpenAI grants resets, free refills you can cash in when you run out mid-cycle. If your usage screen shows "1 available" or "2 available," that is what those are. A reset restores the week and the five-hour window on the spot.
Three things to know:
1. They expire. Mine show expiration dates a few weeks out. Unused resets go away, so check the dates before you assume you have a cushion.
2. You can earn more. The Invite a Friend feature grants a reset when someone you invite actually joins. I have sent so many people to this tool that I would have unlimited tokens if invites carried a link.
3. Resets come before credits, always. Which brings us to the trap.
The Overage Trap
Here is the scenario that costs owners real money. You are on the $200 plan. Two days before your cycle turns over, you run out: weekly window gone, resets spent. The app offers to sell you additional credits so you can keep going.
Those credits are billed at a completely different rate than your subscription. The $200 you pay each month buys usage at a heavily subsidized rate; the top-up credits are closer to raw metered pricing. In those two days, working the way I normally work, I would burn through roughly another $200 per day.
So the decision tree when you hit the wall:
| Situation | Do this |
|---|---|
| Five-hour window empty, week still has room | Wait. It refills in hours. Go run your business. |
| Week empty, resets available | Use a reset. That is what they are for. |
| Week empty, no resets, days left in cycle | Drop to a lighter model and standard speed, or wait. Buy credits only if the work is genuinely worth ~$200 a day. |
| This happens every single month | You outgrew your plan. Upgrade; it is cheaper than topping up. |
What Actually Burns Your Credits
Three dials on the composer decide how fast the meter spins: model, effort, and speed.
Model. The newest 5.6 family (Sol, Terra, Luna) does remarkable work and drinks credits accordingly. The day they launched, I worked with 5.6 Sol for about four hours and went through two resets. For comparison, when Claude Fable came out, arguably the best model on the market right now, I worked it for three and a half hours on a $200 Claude subscription and thought that was fast burn. If your credits are vanishing, drop to 5.5; it is still a great model for most business tasks.
Effort. Light, medium, high, extra high. Save the top settings for problems that have genuinely stumped other models. Running everyday tasks at maximum effort is how you run through a week in a day.
Speed. Standard by default. Fast gives you output sooner and spends tokens faster. Flip it on when you are actively watching something, then flip it back.
Which Plan to Buy
Solo owner starting out: the $100 per month plan. For most of my small business clients, it carries the entire month, and it forces the good habit of measuring before automating.
Owner running real agent operations: $200 per month. When you start building the infrastructure of your business operations around multiple agents and taking on big projects, you graduate naturally. You will know because the $100 plan starts feeling tight in week three.
Teams: this is where owners get burned. The Business tier advertises $25 per seat per month with a five-seat minimum, and $25 of usage will not carry anyone who actually operates agents. My recommendation: the owner and the heavy operators start on individual $100 plans, get the operations in order, and only then consider a pooled plan for the wider team. Pooled billing has its own failure mode, one employee's runaway project draining the whole team's spend, which I cover in the team rollout guide.
And the value math that makes all these numbers small: set up properly, the $100 plan should return the equivalent of a $65,000 per year assistant, at minimum. The $200 plan, at least $100,000 of value, because it can build whatever you envision: apps, internal tools, custom CRMs, custom websites. The gate is never the subscription price. The gate is whether you gave it the access and permissions to actually do the work.
Common Questions
Why does it say I'm out of credits on a $100 plan? You likely exhausted the five-hour window, which refills on its own. Check Usage under your name before assuming the month is gone.
Do resets renew monthly? They are granted with expiration dates and restore your weekly window when used. Watch the dates; they lapse.
Is the $200 plan worth it? Once you run multiple agents and big projects, yes, and it is still a steal against the overage rate and against what the equivalent human output costs.
Can I control costs without upgrading? Yes: 5.5 instead of 5.6, standard speed, effort at light or medium for routine tasks, heavy builds off-peak.
What about my team? Operators on individual $100 plans first, pooled billing only after your agent operations have rules.
Bottom Line
The plan is cheap, the windows refill, and the resets are free. The only expensive thing in ChatGPT Work billing is not understanding it.