Credits
Two balances, a hold before each request, a bill for real usage, refunds, the fuse, and how far one dollar goes.
One credit is one dollar. Angerona shows all prices and balances in USD and tracks them to the millionth, so a sliver of a cent still bills correctly.
Two balances pay for everything
| Bonus balance | Main balance | |
|---|---|---|
| Source | Free daily credits for holding $ANGERONA, $0.50 to $3.00 a day by tier | What you top up with ETH or USDG on Robinhood Chain. See Top up |
| Works on | All models, both modes | All models, both modes |
| Order of use | First | Once the Bonus is gone |
| Lifetime | Fresh at 00:00 UTC each day, leftovers vanish | No expiry |
| Moving or cashing out | Not possible | Not possible. It's prepaid usage: no withdrawals, no wallet-to-wallet moves, no conversion to crypto |
Both figures sit in the app, next to the time your Bonus refreshes.
A request holds money first and pays later
A chat's true cost stays unknown until the model finishes, which is why billing has two stages.
- Hold: just before sending, the gateway blocks the worst-case amount. That is your estimated prompt at the input rate plus the longest allowed answer at the output rate. If both balances combined fall short, nothing goes out and the request is refused.
- Settle: after the last token, real usage is billed and the unused part of the hold returns immediately. Cached prompt tokens cost the model's listed cached-input rate.
Here is a TEE model listed at $1.75 per million input tokens and $5.50 per million output tokens, using the standard 4,096-token answer ceiling.
| Input | Output | Total | |
|---|---|---|---|
| Held | ~3,000 tokens (estimated) = $0.00525 | 4,096 tokens (ceiling) = $0.02253 | $0.02778 |
| Really used | 2,400 tokens = $0.00420 | 500 tokens = $0.00275 | $0.00695 |
| Back in your balance | $0.02083 |
A hold is a ceiling, so it can block a request whose true cost would have fit. The fix is a shorter prompt, a lower answer ceiling, a different model, a top-up or the daily reset. Reasoning models always receive a ceiling of 1,024 tokens or more, since thinking burns tokens as well.
Failures on our side cost you nothing
| Event | Your cost |
|---|---|
| Model error or outage | Zero, full refund |
| Fault inside Angerona | Zero, full refund, flagged apart from "not enough credits" |
| Model can't be served at this moment | Zero |
| You press stop mid-answer | Estimated prompt plus the text streamed so far |
| Model dies mid-answer | Same: estimated prompt plus the text streamed so far |
Images: listed price per image, multiplied by how many you ask for. A failed run comes back in full.
Video: the exact figure displayed for your model, length and resolution. It's held at the start and billed when the file is ready, and a failed or expired job returns all of it.
The fuse protects the treasury
Fees on $ANGERONA trades feed the Angerona treasury, and the treasury pays for every Bonus. Should the Bonus going out start to pass the money coming in, a global fuse multiplies all tiers' daily amounts by one shared factor between 0 and 1.
- All wallets move together, and no single wallet is ever singled out.
- Bonus only. Your Main balance stays untouched.
- New values apply within about 30 seconds.
- A factor of 1 means no effect.
An open cut beats a quiet collapse.
One dollar goes this far
Rough numbers at listed prices, for a message of about 1,500 tokens in and 500 tokens out:
| Item | Mode | Roughly |
|---|---|---|
| DeepSeek V4 Flash messages | TEE | 2,000 |
| Kimi K2.6 messages | TEE | 300 |
| Claude Sonnet 5.5 messages | Anonymous | 125 at official list price, more when a request is served cheaper |
| Images | Anonymous | 10 to 80, by model and size |
Each message in a long chat costs more, because the full conversation travels every time. Reasoning models add thinking tokens on top.
You see the price before you send
- Up front: the picker displays the present price per million tokens (input, output, cached input), per image or per video.
- Anonymous text: served from a secondary compute market, where rates tend to sit under the official list price. The cheapest available offer for the model is tried first, and you owe what that request truly cost, capped at the official list price.
- TEE: the official list price, exactly, on the tokens consumed.
- Images and video: the listed price.
- One price list: tiers resize your daily Bonus and leave rates alone.
- No subscriptions: usage only, Bonus first, Main after.
- Moderation: screening image and video prompts costs you nothing.