Cost of 1,000 six-second AI video clips by model

1,000 six-second clips cost $75 on Grok Imagine Video 1.5, $450 on MiniMax H3 at 768p, $750 on Wan 3.0 at 720p and $1,260 on Kling 3 with audio on. The total is 6,000 generated seconds times the per-second rate, so the model choice is the whole bill.

Batch totals

Each row multiplies the Sume billable rate (provider list times 1.25) by 6,000 seconds. Every model shown accepts a 6-second request.

Do not submit 1,000 at once

Generation admission limits how many jobs a workspace holds. Accepted capacity is 6 on Free, 24 on Pro, 48 on Startup and 120 on Scale and Enterprise; a request beyond it returns 429 queue_full. The generation_limits field shows the values for your workspace.

Submit in waves that stay under the accepted cap and wait for completions before the next wave. The sketch below shows the loop; replace the pause with polling the job status.

import time

def submit_in_waves(prompts, submit, accepted_cap, pause=30):
    """submit(prompt) -> job id; stops at the accepted-job cap per wave."""
    ids = []
    for i in range(0, len(prompts), accepted_cap):
        wave = prompts[i:i + accepted_cap]
        ids += [submit(p) for p in wave]
        time.sleep(pause)  # replace with polling until the wave is done
    return ids

Wallet check

Each accepted job reserves its estimate at submit, so a full wave holds money before any clip finishes. On Pro, a wave of 24 Wan 720p six-second clips holds $18.00. If the wallet cannot cover a job, it returns 402 insufficient_credits before provider work starts.

• Wave size times per-clip cost is the most money held at once.
• Failed jobs release their reserve; successes capture it.

Choosing a row for a batch

For a bulk job, choose by the finished use. Thumbnails, social loops and test variants rarely need 1080p, and the cheapest rows cut the bill by an order of magnitude. If the batch will be reviewed by a person, run a 20-clip pilot on two or three rows first: 20 pilot clips on Wan 720p cost $15.00 and tell you whether the $750 total is justified.

Also consider the audio. Kling 3 with audio on, Gemini Omni Flash 1.1 and MiniMax H3 Max generate a native track; check generate_audio in the catalog for the row you pick before you budget separate music or voice.

Limits

The tables show the model amount. A workspace Agent Fee line can be added on top, so treat the job's own usage cost as the truth and check the live catalog before you budget; limits and rows change. Concurrency and queue values are plan-dependent and the static table above can lag the live generation_limits response.