Comparison · Pricing

Video Encoding API Pricing, Explained

Encoding APIs price by the minute, by GPU-hour, or by subscription tier — and the fine print (egress, source fetch, queue time) moves the real number more than the headline rate. Here is how to compare honestly.

PricingVideo encodingAPIs

Models

The three pricing families

ModelExample shapeGood forWatch out for
Per minute/second$0.004/min of inputVOD, variable volumeInput or output minutes? read the fine print
Per GPU-hour$/hour of rented transcode capacitySustained heavy volumeIdle time, minimum commitments
Per-tier subscriptionMonthly plan + included minutesPredictable small volumeOverage rates, feature gating

Hidden costs that change the bill

Note: Two vendors can quote the same 'per minute' price and differ 40% on a real job. Ask for a bill simulation on YOUR workload before comparing.
  • Egress: what it costs to get the file back out (or is it included?)
  • Source fetch: does the API pull from your bucket for free?
  • Per-resolution multiples: 4K often bills 4× the 1080p minute
  • Queue/priority: are fast lanes charged extra?
  • Failed jobs: refunded or billed?
  • Storage: keep it in your bucket, not theirs

Estimating your real cost

Real cost = (input minutes × per-resolution multiplier × codec multiplier) + egress + source-fetch. Build this on your actual catalog: a 10,000-hour 1080p H.264 library encoded to H.264 + HEVC + AV1 ladders is a very different number from a monthly 500-minute social upload.

Our model, plainly

Line itemmlslabs
UnitPer minute of input video
Failed jobs$0 — auto-refunded
EgressWrite to your object storage; no egress charge from us
ResolutionsSingle rate per minute; volume discounts
BatchOne queue, per-file callbacks, same rate

FAQ

Common questions

Why do encoding prices vary so much?

Different unit definitions (input vs output minutes, per-resolution multiples, included vs billed egress) make headline rates incomparable. Normalize on your real workload.

Is per-minute or per-GPU-hour cheaper?

At low volume, per-minute wins (no idle cost). At very high sustained volume, owned/rented GPU capacity can beat it — model your utilization.

Do failed jobs cost money?

Not with us — credits auto-refund. Some vendors bill failed jobs; check before committing.