The quick pricing answer
The vgenv MiniMax H3 API starts at $0.0015 per generated second for Fast 480p. Fast 768p is $0.003 per second, while Quality is $0.006 per second at 480p and $0.012 per second at 768p. There is no monthly API subscription: each accepted request receives a server-authoritative price snapshot before generation begins.
These are vgenv prices for the public model identifier minimax/h3. vgenv is an independent API service, not MiniMax's first-party API, so compare the complete request, resolution, duration, queue behavior, and output policy rather than assuming model identifiers or prices are interchangeable.
| Tier | 480p | 768p | Typical use |
|---|---|---|---|
| Fast | $0.0015/s | $0.003/s | Prototypes, drafts, and batch creative testing |
| Quality | $0.006/s | $0.012/s | Higher-quality production outputs |
Cost examples for 5, 10, and 15 second videos
Use a simple formula: generated duration × the selected resolution and tier rate. Input preparation, queue time, and status polling do not add generated seconds. Always use the price returned by POST /v2/videos as the final quote for that job.
| Configuration | 5 seconds | 10 seconds | 15 seconds |
|---|---|---|---|
| Fast 480p | $0.0075 | $0.015 | $0.0225 |
| Fast 768p | $0.015 | $0.030 | $0.045 |
| Quality 480p | $0.030 | $0.060 | $0.090 |
| Quality 768p | $0.060 | $0.120 | $0.180 |
How quote snapshots and failures work
When a request is accepted, vgenv calculates the exact quote from the published configuration and reserves that amount from the Project balance. The accepted Video resource contains the immutable price snapshot, so a later catalog update cannot silently change the cost of an existing job.
A successful job settles the reserved amount. A platform failure, an unsupported execution path detected after acceptance, or a queued cancellation releases the reservation. Your integration should still treat the returned Video status and price fields as authoritative and should not calculate ledger changes in client code.
- Log the Video ID and returned price snapshot for reconciliation.
- Reuse an Idempotency-Key only for the same logical request.
- Handle failed and cancelled as terminal states.
- Download successful output before its signed URL expires.
How to compare MiniMax H3 API providers fairly
A low headline rate is useful only when the supported configuration matches your workload. Compare the same output duration, resolution, input mode, quality tier, concurrency policy, failure policy, and storage behavior. Also check whether the API exposes a stable asynchronous resource, idempotent creation, cancellation, and a final server-side quote.
For a real procurement test, send a fixed prompt set through each candidate, record successful cost per usable output, and review motion, identity consistency, prompt adherence, latency, and retry rate. Keep first-party and independent-provider keys in separate server-side secret scopes because their endpoints and request schemas are not drop-in compatible.
Practical ways to control generation spend
Start evaluation runs with Fast 480p, promote only selected prompts to 768p or Quality, and cap active jobs per Project. Idempotency prevents accidental duplicate creation, while exponential polling backoff reduces status traffic without delaying the generation itself.
- Run prompt and composition experiments at Fast 480p.
- Use deterministic Idempotency-Key values for retried create calls.
- Set per-user or per-workspace generation limits in your application.
- Store the returned quote and output metadata for cost attribution.