Models & credits
Credits
How the Bonus and Main balances work, the daily Bonus renewal, how each request is reserved and then charged for actual usage, refunds, the fuse, and what a dollar buys.
On this page
Credits are dollars. Every price and balance in Tacita is shown in US dollars and counted in millionths of a dollar, so even a fraction of a cent is charged exactly.
You pay only for what you use. Each model has a transparent price per token, per image or per video, and the pricing is the same for everyone. There are no subscriptions.
Two balances#
| Bonus balance | Main balance | |
|---|---|---|
| What it is | Free daily credits for holding the Tacita token. Your tier sets the amount: $0.10 to $1.00 a day. | Your own top-ups in ETH or USDG. See Top up. |
| What it works on | Every model, in both modes | Every model, in both modes |
| Charged | First | After your Bonus balance runs out |
| Renewal and expiry | Renewed every day at 00:00 UTC. Unused Bonus doesn't roll over. | Never expires |
| Can be moved or cashed out | No | No. It's prepaid usage, and can't currently be withdrawn, transferred to another wallet or converted back to crypto. |
Every charge comes from your Bonus balance first, then from your Main balance. The app shows both balances and when your Bonus balance next renews.
Reserve, then settle#
You can't know the exact cost of a chat request before the model answers, so every request works in two steps.
- Reserve. Before your request goes out, the gateway sets aside the most it could cost: your estimated prompt at the input price, plus the maximum answer length at the output price. If your Bonus and Main balances together can't cover that amount, the request is refused before anything is sent.
- Settle. When the answer is done, you're charged for actual usage, and the rest of the reservation goes straight back to your balances. In TEE mode that's exactly the model's official list price for the tokens used. In Anonymous mode you're billed the actual price the request was served at, which is never more than the official list price (see How prices work). Prompt tokens served from cache are charged at the model's listed cached-input rate.
Example. Take a TEE model with an official list price of $1.75 per million input tokens and $5.50 per million output tokens, with the default maximum answer of 4,096 tokens.
| Input | Output | Total | |
|---|---|---|---|
| Reserved | ~3,000 tokens (estimated) = $0.00525 | 4,096 tokens (maximum) = $0.02253 | $0.02778 |
| Actually used | 2,400 tokens = $0.00420 | 500 tokens = $0.00275 | $0.00695 |
| Returned to you | $0.02083 |
Because the reservation is an upper bound, a request can be refused even when the actual usage would have fit. If that happens, shorten the prompt or the maximum answer length, pick another model, top up, or wait for the daily reset. Reasoning models always get a maximum of at least 1,024 tokens, because their thinking uses tokens too.
When you aren't charged#
| What happened | What you pay |
|---|---|
| The model fails or is unavailable | Nothing. Full refund. |
| Something fails on Tacita's side | Nothing. Full refund. It's reported separately from "not enough credits". |
| The model can't be served right now | Nothing |
| You stop the answer mid-stream | The estimated prompt plus the part of the answer already streamed |
| The model fails mid-stream | The same rule: the estimated prompt plus what was streamed |
Images are charged the listed price per image, times the number of images. If generation fails, you get a full refund.
Video is charged the exact price shown for the model, length and resolution you pick. That amount is held when the job starts and charged when the video is ready. If the job fails or expires, the hold is refunded in full. A video can't be stopped once it is rendering.
The fuse#
Bonus balances are paid for by the Tacita treasury, which is funded by fees on Tacita token trades. If the Bonus handed out starts to exceed what the treasury takes in, a global fuse scales every tier's daily Bonus down by the same factor, somewhere between 0 and 1.
- It applies to everyone at once, and never to a single wallet.
- It affects the Bonus balance only. Your Main balance is never scaled.
- A change takes effect within about 30 seconds.
At 1, the fuse changes nothing. We'd rather turn the Bonus down openly than run out of money quietly.
What $1 buys#
Approximate figures at current prices, for a typical message of about 1,500 tokens in and 500 tokens out:
| What | Mode | About |
|---|---|---|
| DeepSeek V4 Flash messages | TEE | 2,000 |
| Kimi K2.6 messages | TEE | 300 |
| Claude Sonnet 5.5 messages | Anonymous | 125 at the official list price, or more, depending on the price each request is served at |
| Images | Anonymous | 10 to 80, depending on the model and size |
Long conversations cost more per message, because the whole conversation is sent each time. Reasoning models also spend tokens on thinking. The model picker always shows current prices.
How prices work#
- Transparent. Each model's price per million tokens (input, output and cached input), per image or per video is shown in the picker before you send.
- Anonymous mode (text) runs on a secondary compute market. Prices are typically below each model's official list price, and every request is routed to the lowest available price for that model first. You're billed the actual price the request was served at, and never more than the official list price.
- TEE mode: you're charged exactly the model's official list price for actual usage.
- Images and video: you're charged the listed price.
- The same for everyone. Your tier changes the size of your daily Bonus balance, never the price.
- No subscriptions. You pay only for what you use, from your Bonus balance first, then your Main balance.
- Prices can change as models change. The picker always shows the current price.
The moderation check on image and video prompts is free for you.