Speech-to-text pricing

78 STT model(s) from 18 provider(s). Priced from input_audio_kseconds, US dollars per 1,000 seconds of input audio, and shown converted to a per-minute rate for comparison. Vendors quote in whatever unit they prefer (per minute, per second, per hour), so the figure here is a conversion, not a quote. Pick a vendor in the sidebar to see that vendor alone. Streaming and batch ship as separate rows, as do language tiers, because vendors price them separately. 2 further STT model(s) bill audio as tokens rather than by the second and are listed on their vendor page: stt-async, stt-rt.

• api the vendor publishes this rate at a machine-readable endpoint, so it is re-read and compared automatically rather than by a human reading a pricing page.
• estimated this rate is not the meter the vendor bills on, so it will not reconcile against an invoice. Either the model is billed in another unit and this figure is the vendor restating it, or the vendor publishes only a floor (“starting at”) or a range. Useful for comparison and capacity planning, not for billing.

​

Via LiveKit Inference

What the same model costs going direct to the vendor versus through the gateway. Scale is the discounted tier. 4 further LiveKit STT model(s) have no direct-vendor entry to compare against. What verified, imported, and seed mean is on How Fresh?.