One minute of AI video at Higgsfield's listed per-second rates
Higgsfield's API post lists six per-second video rates, from $0.042 to $0.20. The cost of 30 seconds and one minute for each, and what a three-take retry does.

At the rates on Higgsfield's API post, one minute of generated video costs between $2.52 for Kling 2.5 and $12.00 for Wan 3.0, before retries. The Higgsfield API blog lists six per-second prices, and multiplying by 60 gives the cost of a minute of output.
Rates and totals
Rates are as listed on the post on 2026-10-03. The post's own comparison page lists a different Seedance 2.5 figure, covered in a separate post on this blog, so treat all of these as list rates to confirm.
| Model | Per second | 30 seconds | 60 seconds |
|---|---|---|---|
| Kling 2.5 | $0.042 | $1.26 | $2.52 |
| Kling 2.6 | $0.07 | $2.10 | $4.20 |
| Seedance 2.5 | $0.0738 | $2.21 | $4.43 |
| Kling 3.0 | $0.112 | $3.36 | $6.72 |
| LTX 2.5 Pro | $0.17 | $5.10 | $10.20 |
| Wan 3.0 | $0.20 | $6.00 | $12.00 |
Add the retake multiplier
Nobody ships the first take. If you keep one take in three, the cost of a finished minute is three times the table, so Seedance 2.5 becomes about $13.28 and Wan 3.0 $36.00 per finished minute. The three-to-one ratio is an assumption for illustration. Use your own keep rate.
rates = {"Kling 2.5": 0.042, "Kling 2.6": 0.07, "Seedance 2.5": 0.0738, "Kling 3.0": 0.112, "LTX 2.5 Pro": 0.17, "Wan 3.0": 0.20}
takes_per_keeper = 3 # your own keep rate, not a vendor number
for model, per_second in rates.items():
print(f"{model}: ${per_second * 60:.2f} per generated minute, ${per_second * 60 * takes_per_keeper:.2f} per finished minute")Cheap per second is not cheap per result
The lowest per-second rate is not automatically the lowest cost per approved shot. A cheaper model that needs five takes can cost more than a pricier one that needs two. Track your own keep rate per model and multiply, as the code above does.
Length limits also change the math: a model capped at a short clip needs more jobs to fill a minute, and each job has its own failure risk. Compare on cost per approved minute, which includes takes and joins, not only the listed rate.
- Measure keep rate per model on your own prompts.
- Count jobs needed for the target length.
- Compare cost per approved minute.
- Re-measure when a model version changes.
What does not transfer
List rates on one vendor do not carry over to another. On Sume the video docs say a job reserves provider list times 1.25 at submit and reports the billable amount as usage.cost, and the admission docs say failed jobs release or refund the reservation. Price your own plan from pricing_skus and a pilot run, not from another vendor's table.
Sources
Related posts
More in Pricing
- Runway credits to seconds: Gen-4.5 is $0.096 to $0.23 per second
Runway's pricing page lists Gen-4.5 at 60 credits per 5 s and Seedance 2.0 Pro 1080p at 160 per 4 s. Converted to seconds and dollars across three plans.
- Veo 3.1 Lite at 720p: eight drafts cost the same as one Standard clip
At Google's listed rates, one 8-second Veo 3.1 Standard clip costs $3.20 and eight Lite 720p clips cost the same. Here is the draft-then-final math.
- Video API billing terms checklist: expiry, refunds, storage
Higgsfield's API post lists pay-as-you-go USD, no charge for failed requests, funds expiring in one year and files kept 7 days. A buyer checklist built on it.
- How Sume pricing works: plans, one wallet, published model rates
Sume plans set access and concurrency. Usage draws from one prepaid wallet at each model's published USD rate, for generation, the Agent, Formats, and the API.
Written by Sume