API Plan Finder
Prices and limits match the providers' official pagesChecked 2026-11-11

Speech-to-text API prices: distribution over 10 tracked plans

Beta: statistics over the plans Plan Finder tracks and has checked, not the whole market. At 100 audio hours a month, checked 2026-11-11 against the official pages.

What is tracked

5 providers, 5 products and 10 plans. 10 of 10 plans state a monthly cost at this volume; 0 cannot serve it and 0 give no cost for it; neither group is counted as cheap or expensive.

Monthly cost at 100 audio hours

Monthly cost in USD
Min25th percentileMedian75th percentileMaxPlans costed
$15.00$19.50$30.90$83.55$142.5610 of 10

min and max are the lowest and highest; p25, median and p75 are linear interpolation between the closest ranks (type 7), in USD rounded to cents.

Price per unit in USD (price per audio hour)
Min25th percentileMedian75th percentileMaxPlans costed
$0.15$0.20$0.31$0.84$1.4310 of 10

min and max are the lowest and highest; p25, median and p75 are linear interpolation between the closest ranks (type 7), in USD rounded to cents.

The 3 cheapest plans

  1. AssemblyAI Pay as you go, pre-recorded (Universal-2): $15.00 a month
  2. AssemblyAI Pay as you go, streaming (Universal-Streaming): $15.00 a month
  3. Google Cloud V2 API Dynamic Batch (pay as you go): $18.00 a month

Free tier

1 of 10 plans have a free allowance that repeats every month; 5 do not; 4 are not stated on the official pages.

Conditions across 10 plans

ConditionYesNoNot stated (unknown)Of plans
Speaker diarization is stated51410
Billed by the second with no minimum per request is stated50510
Korean is supported70310
Streaming (real-time) is stated30710
Customer content is not used to train the vendor's models (by default)60410
The provider holds a SOC 2 report30710

Unknown means the official pages do not say; it is not counted as no. Computed by code from the catalog with no model: each plan cost is the engine's monthly cost at the volume (search, no budget); condition counts are the engine's verdicts per plan (yes = met, no = not met, unknown = the official pages do not say); a provider counts as yes, no or unknown only when all its plans do, else mixed. The same input gives the same output.