Munsit-1 scores 26.68 average word error rate across six independent Arabic datasets — 27% fewer errors than Whisper. Read the methodology, then start with one curl.
Send Arabic audio in, or get Arabic speech out. Flip to API call and you're looking at the precise request that produced what you just saw — no hidden magic, no rewriting.
No dialect field — the model works it out.
Same base URL, same header, same shape. Change the path, change what you get — transcription, speech, streaming, cloning. Nothing else moves.
Eighteen endpoints and five agent plugins — every one of them live today, and most of them currently reachable from nowhere on the site. This is the section that tells a developer the product is bigger than a transcribe call.
munsit-1 · #1 on Arabic ASR
faseeh-v1 · 16 voices
nothing else does these
on top of any transcript
clean it before you use it
drop-in plugins
Arabic isn't one language. A Moroccan and a Kuwaiti use different words for the same sentence, and general-purpose models flatten both into wrong MSA. Pick a dialect, play the clip, read what came back.
Notice what isn't in the call: a dialect parameter. You don't tell Munsit which dialect you're sending — it works that out. The same request handles all 25, so nothing needs re-integrating when your users change.
Start free with 10,000 credits — no card. Every tier's price is on this page, and purchased credits never expire.
Growth ($250/mo · 10M credits · uptime SLA) and Scale ($800/mo · 50M credits · dedicated infra) also available. See all plans → · Annual billing saves 20%.
One key, 10,000 free credits, every endpoint on this page. You'll have a transcript back before this tab finishes loading twice.
no credit card · no sales call · unused credits never expire