Three fine-tuned models. One endpoint. Pay per minute — no contracts, no minimums.
curl -X POST https://api.kohnai.ai/v1/transcribe \ -H "Authorization: Bearer ak_live_7f3a9b2e..." \ -H "Content-Type: application/json" \ -d '{ "model": "whisper", "audio_url": "https://storage.example.com/shiur-42.mp3" }' # → 200 OK { "text": "אין יענעם יאר האבן מיר זיך אלע צוזאמענגעקליבן...", "credits_used": 6, "remaining_credits": 994 }
Purpose-built for Yiddish speech. Pick by accuracy, cost, or recording type.
Highest accuracy. Fine-tuned Gemini on Yiddish via Vertex AI. Handles archival recordings, heavy accents, and degraded audio.
Best for: archival digitization, Rebbe recordings, lectures
Fast and affordable. GPU-accelerated Whisper fine-tuned on Yiddish. Best cost-to-accuracy ratio for bulk processing pipelines.
Best for: bulk batches, podcasts, shiurim archives
Purpose-built for modern and broadcast Yiddish. Strong on casual speech, interviews, and media with contemporary vocabulary.
Best for: interviews, news broadcasts, media
One REST endpoint. No SDK required. No email confirmation to wait for.
Create an account — no email confirmation required. You'll get 100 free credits immediately. API keys start with ak_live_.
$10 gets you 1,000 credits. Credits never expire. Scale from one file to millions — same API, same pricing, no lock-in.
Send an audio URL to /v1/transcribe with your model choice. Yiddish text back in seconds.
1 credit = $0.01. All models. No subscriptions. Credits roll over forever.