Changelog
Model additions, pricing changes and API changes, dated. Anything that changes what you are charged appears here.
- Platform
Durable job execution
Generations now survive the process that started them. A job is dispatched after the response is sent, then advanced one poll at a time by a scheduler, so a 30-second generation is no longer bounded by how long a single server invocation may live. Every step is idempotent: a lost or duplicated run costs latency, never credits.
- Pricing
Per-job credit ceiling raised to 12,000
The original 5,000-credit ceiling would have refused a legitimate 30-second 1080p generation on Seedance 2.5, which prices around 7,200 credits. The ceiling now sits at 12,000 — still low enough to refuse a 30-second 4K job, which prices near 17,000 and is almost always a mistake. Plan gates are unchanged.
- Model
Seedance 2.5 added, with claims marked as claims
Added Seedance 2.5 — 30 seconds in a single pass, up to 50 references. Its 4K output, reference ceiling, native audio and the region-edit / R2V / 3D-blockout controls are ByteDance’s own announced figures and are labelled claimed until we confirm them against the provider API. No independent benchmarks for 2.5 existed when this page was written.
- Pricing
Cost calculator publishes our upstream cost
The calculator shows the retail credit price and the underlying provider cost per second side by side, with the margin multiplier stated. Rates currently derive from published vendor pricing; they are replaced by our own measurements when the benchmark run lands, and this page will say when.
- API
Public API v1
/api/v1 opened: create, fetch, list and cancel generations, read the model registry with live credit rates, and read your balance. Idempotency-Key honoured on create, rate-limit headers on every response, and a stable error shape with documented codes.