Pricing & Credits
How credits are charged, when they are refunded, and how to read your balance
Pricing & Credits
The API is pay-per-generation in credits. One task = one charge, regardless of how long the model takes.
Cost per model
| Model | Credits per generation |
|---|---|
tripo | 8 |
tripo-text | 8 |
trellis2 | 10 |
trellis2-multiview | 12 |
hunyuan3d | 12 |
tripo-multiview | 15 |
Per-model capabilities and limits live in Available Models.
GET /api/v1/models returns the same creditsCost values, so you can read
prices at runtime instead of hardcoding them.
When you are charged
Credits are deducted at task creation, not at completion. That means:
POST /api/v1/3d/generationssucceeds → credits are already gone, even if the generation later fails.- A failed generation is refunded. The task response then carries
usage.credits_refunded. - A request rejected before a task is created —
400,401,402,403,429— costs nothing.
Deduction is atomic — it only applies when the balance covers the cost. If you
fire ten generations in parallel with credits for five, five succeed and the
rest get 402 without being charged. The balance can never go negative.
Reading your balance
Every successful response includes the post-transaction balance:
{
"id": "task_abc123",
"status": "completed",
"data": { "model_url": "https://r2.example.com/model.glb", "format": "glb", "model": "trellis2" },
"usage": {
"credits_consumed": 10,
"credits_remaining": 490
}
}| Field | Present when |
|---|---|
credits_consumed | Task completed successfully |
credits_remaining | Any response that carries a usage block |
credits_refunded | A failed task was refunded |
Treat credits_remaining as the source of truth. Do not subtract creditsCost
locally to predict a balance — refunds, retries and concurrent tasks make local
arithmetic wrong quickly.
Out of credits
HTTP/2 402{
"error": {
"message": "Insufficient credits",
"type": "credits_error",
"code": "insufficient_credits",
"param": null
}
}402 is not retryable without a top-up. Check the balance before large batches:
run one cheap generation and read credits_remaining, or read the balance in
your dashboard billing page.
Estimating a batch
total credits ≈ Σ (credits of chosen model × number of generations)A 100-image batch on trellis2 is 100 × 10 = 1000 credits. If you expect
retries — bad inputs, provider timeouts — provision 10–15% extra rather than
discovering the ceiling mid-batch.
Credits are shared between the web Studio and the API. Generations you run in the browser spend from the same balance, so a busy day in the UI reduces what your integration has available.