Pricing
One balance, no subscription, and the number on screen before you spend it.
How you get tokens
Today we grant them by hand — ask us and say roughly what you want to run. There is no card payment and nothing on this site takes money. Tokens do not expire and there is no monthly charge to forget about.
What things cost
Prices below are in tokens and already doubled — this is what you pay, not what we pay. A chat message costs a fraction of a token. Pictures and video are priced by size and length, and what each setting actually cost is further down.
| Model | Good for | Per message |
|---|---|---|
| AiHub AIThe free assistant, always on | The free assistant, always on | Free |
| Claude Haiku 4.5Fastest and cheapest | Fastest and cheapest | ~0.06 |
| Gemini 2.5 FlashFast and cheap | Fast and cheap | ~0.06 |
| Gemini 3 FlashFast multimodal | Fast multimodal | ~0.08 |
| Grok 4.3Lighter, quicker Grok | Lighter, quicker Grok | ~0.08 |
| GPT 5.2Balanced everyday reasoning | Balanced everyday reasoning | ~0.1 |
| Gemini 3.5 FlashFast, newer generation | Fast, newer generation | ~0.1 |
Pictures and video quote first
They cost tens or hundreds of tokens and take minutes, and once a generation starts the provider bills us whatever you make of the result. So the app quotes you before it runs — the exact number, the settings it depends on, and what your balance will be afterwards — and nothing starts until you say yes.
What a picture and a clip actually cost
Nothing here is worked out. Each row is one real generation whose reported cost was recorded, because no provider publishes what a setting does to the bill. That is also why these figures carry no tilde while the ones on the model list do.
The two pages are not disagreeing; they are quoting the same models at different settings. The model list prices each one at its own defaults — the smallest size on pictures, 720p at five seconds on video — so every picture model’s first row here is the figure you will find there, and every video figure there sits above the rows here, which are three- and four-second runs.
| Model | Settings | Took | Tokens |
|---|---|---|---|
| Nano Banana 2 LiteTook 11s | Only one size | 11s | 8 |
| GPT Image 1.5Took 1m 32s | Medium quality | 1m 32s | 8 |
| GPT Image 1.5Took 5m 47s | High quality | 5m 47s | 44 |
| GPT Image 2Took 1m 9s | 1K | 1m 9s | 12 |
| GPT Image 2Took 1m 14s | 2K | 1m 14s | 20 |
| GPT Image 2Took 1m 26s | 4K | 1m 26s | 32 |
| GPT Image 2 EditTook 1m 12s | 1K | 1m 12s | 12 |
| GPT Image 2 EditTook 1m 40s | 2K | 1m 40s | 20 |
| GPT Image 2 EditTook 1m | 4K | 1m | 32 |
| Nano Banana 2Took 25s | 1K | 25s | 16 |
| Nano Banana 2Took 1m 3s | 2K | 1m 3s | 24 |
| Nano Banana 2Took 59s | 4K | 59s | 36 |
| Nano Banana ProTook 44s | 1K | 44s | 36 |
| Nano Banana ProTook 1m 2s | 2K | 1m 2s | 36 |
| Nano Banana ProTook 2m 10s | 4K | 2m 10s | 48 |
| Model | Settings | Took | Tokens |
|---|---|---|---|
| Seedance 1.5 ProTook 4m 59s | 720p4s | 4m 59s | 28 |
| Seedance 1.5 ProTook 1m 58s | 1080p4s | 1m 58s | 60 |
| Seedance 2 MiniTook 1m 41s | 480p4s | 1m 41s | 30.4 |
| Seedance 2 MiniTook 2m 52s | 720p4s | 2m 52s | 65.6 |
| Kling V3 TurboTook 36s | 720p3s | 36s | 108 |
| Kling V3 TurboTook 58s | 1080p3s | 58s | 135 |
| Seedance 2 | 480p4s | Not timed | 152 |
| Seedance 2.5Took 3m 48s | 480p4s | 3m 48s | 224 |
What the free assistant costs
Nothing, including when a paid model is answering for it behind the scenes. That cost is ours, and it is what keeps the free tier from being a coin flip when the free models are busy.
When something fails
A request that never reached a provider is never charged. One that failed midway is charged for the part that actually ran, doubled — usually a fraction of the quote. The refunds page spells out each case.