AssemblyAI vs Tunk

AssemblyAI

The transcription API to reach for when accuracy and per-hour pricing matter more than a user interface, but the headline rate is a base rate and every label you actually want adds to it.

Pros

• +$50 of credits at signup with no card required
• +Per-hour rates are published plainly and undercut most incumbents
• +Add-ons cover diarisation, redaction, summarisation and medical vocabulary
• +Voice Agent API is one flat $0.075/min covering model and speech too

Cons

• −No product surface at all — everything is code
• −Add-ons stack, so the real rate usually lands well above $0.15/hr
• −Streaming diarisation costs six times the async equivalent
• −No volume discounts published on the pricing page

Tunk

A small Indian voice-AI vendor covering both halves of the stack — recognition and agents — with playgrounds you can poke at before wiring anything up. What we could not establish is the thing that decides whether it is worth trying: the rate. Nothing on the site rendered a price for us, and there is little third-party coverage to check it against.

Pros

• +Covers both speech-to-text and voice agent building from one account
• +Interactive playgrounds for both products, not just API documentation
• +Names a real legal entity and registered Indian address on the site

Cons

• −Pricing page renders no figures, so the rate is unknown until you sign up
• −No published accuracy or latency benchmarks
• −Small vendor with thin third-party coverage to corroborate its claims

Which has a public API, AssemblyAI or Tunk?

Both ship a public API, so either can be built on; the fact table above shows the platforms each one runs on.