OpenCode Go Usage Screen: What the Numbers Actually Mean (and Why It Looks Like You're Almost Out When You're Not)
This is a plain-words guide to the OpenCode Go usage page — so you can tell what you've really spent, what's still available, and which numbers on that screen to ignore.
Independent write-up. Not affiliated with Anomaly/OpenCode. Verified against a live subscription, September 2026. Limits change — check https://opencode.ai/docs/go for current values.
- You pay $10/month. You get about $325/month of model usage, split up.
- The screen shows big percentages that make it look like you're almost out when you're not.
- The "Total 70%" number is made up. They added up percentages that each have a different limit. Adding those gives a number that means nothing.
- Real example from a working subscription: $21 actually used out of $325 available = 6.5%. The screen said 70.2%. That's ~11x scarier than reality.
There are TWO separate limit systems on the same page. The screen never tells you that.
1. The bars at the top = one shared wallet for everything.
- 5-hour bar — shared wallet, refills every 5 hours. Only matters if you're running something huge.
- Weekly bar — same wallet, bigger. Around $30.
- Monthly bar — the one that actually matters. Around $60. Resets monthly.
2. The model table = each model has its own private wallet.
One model can be almost empty while all the others are still full. That's what happened in the example above: one model at 60% of ITS OWN small limit, everything else barely touched.
| Screen says | What it means for you |
|---|---|
| "5-hour Usage 0.2%" | Shared wallet, refills 4x a day. Ignore it. |
| "Weekly Usage 14.2%" | Same shared wallet, ~$30 worth. |
| "Monthly Usage 70.2%" | The real one. ~$60 worth. Watch this. |
| Model table % | Each model's own private wallet. They drain separately. |
| "Total 70.2%" | Made-up number. They summed percentages with different limits. Ignore it. |
| "Resets in 4h 55m" | Refill time. Not a deadline. Don't panic. |
| "Use your available balance after reaching the usage limits" | They want to bill you extra after the caps run out. Optional — decide yourself, don't enable by default. |
Say it simply:
- Big percentages instead of dollars. "$21 spent, $304 free" feels like winning. "70% used" feels like losing. Same facts — they picked the one that makes you anxious.
- Countdown timers everywhere. That's the phone-game energy-refill trick. They want you checking in on their page.
- The deal only pays off if you use ALL the models. Most people use one model heavily, the rest of their $325 expires unused. They keep the rest. That's the business model.
- Nobody explains the words. When only the company understands the bill, the company wins.
To be fair: the deal itself is good. $10 for $325 of usage is real value. The problem is only the screen — it hides a good deal behind a scary fake number.
- Match model to job. Big thinking → the model with the biggest cap (often 99% unused). Routine work → the cheap models. It's your money spread across models — use all of it, not just one.
- Freeze the model that's near its cap until the monthly reset.
- Agent loops cost more than you think. Every tool call is a new billed request. 200 tool calls = 200 requests. Route accordingly.
- Check the 5-hour bar before long sessions. It refills 4x a day, so start big runs right after a reset.
Check your real windows from the terminal (no browser, no scraping):
curl -s -H "Authorization: Bearer $OPENCODE_API_KEY" \
https://opencode.ai/zen/go/v1/usageThat one command tells you more truth than their whole console page.
If this saved you an evening of confusion, pass it on.