Skip to main content

Comparison

soniox/stt-rt-v4 and soniox/stt-rt-v6 are listed for AI Assistants, WebSocket Streaming, and the Voice API; soniox/stt-rt-v5 is listed for AI Assistants. Term biasing (context) and language_hints are AI Assistants settings supported by stt-rt-v4, stt-rt-v5, and stt-rt-v6 — they aren’t gated to a version.Soniox has deprecated stt-rt-v5. From December 15, 2026, Soniox serves stt-rt-v5 and stt-rt-v4 (an alias of v5) with v6. v6 writes prices as words (for example, “15 dollars and 20 cents” rather than “$15.20”) and includes hyphens in codes and IDs only when they are spoken.

Engine Details

The default WebSocket engine. Best English accuracy and the richest feature set. For REST, you must explicitly set model="deepgram/nova-3" — the REST default is openai/whisper-large-v3-turbo.Models:
  • nova-3 — Latest and most accurate. Supports diarization, word-level timestamps, smart formatting, numerals, and punctuation via model_config. Use this unless you need the lowest possible latency.
  • nova-3-medical — Nova-3 tuned for clinical vocabulary (medication names, dosages, medical terminology). Same feature set as nova-3, but English only. Available over WebSocket streaming as model=nova-3-medical with transcription_engine=Deepgram (non-English language values are coerced to en) and on the REST API as deepgram/nova-3-medical.
  • nova-2 — Previous generation. Still supported but nova-3 is better in all benchmarks.
  • nova-2-medical — Previous-generation clinical model, English only. REST only, as deepgram/nova-2-medical.
  • flux — Purpose-built for voice agents. Lowest latency with built-in end-of-turn detection — tells you when the speaker has finished so your agent can respond. English only. WebSocket only.
  • flux-multi — Multilingual Flux with code-switching. Same latency and end-of-turn detection as flux. WebSocket only.
Languages: Coverage differs by surface — the REST batch models load more per-language models than the streaming ones, so deepgram/nova-3 reaches ~49 languages on REST and fewer over WebSocket, while deepgram/nova-2 is ~33 on both. Nova-3 supports multi mode (10 languages with code-switching). Flux-multi supports English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian, and Dutch, plus auto. See Deepgram languages.

How to Choose

Need the highest accuracy for English? → Deepgram nova-3 — best WER (word error rate) across all English variants. Building a voice agent that needs to know when the user stopped talking? → Deepgram flux — lowest latency with built-in end-of-turn detection. Use flux-multi for multilingual agents. Need to transcribe files in 50+ languages? → Telnyx openai/whisper-large-v3-turbo via REST API. Need diarization (who said what)? → Deepgram nova-3 with model_config.diarize: true. Need separate transcripts per channel of a stereo recording? → Deepgram nova-3 with model_config.multichannel: true (billed per channel). See Multichannel. Need broad accent/dialect support? → Azure azure/fast — strong coverage across regional accents. Need Grok STT for real-time calls? → xAI xai/grok-stt via WebSocket or Voice API. Need low-latency streaming for voice agents? → AssemblyAI assemblyai/universal-3-6-pro via WebSocket or Voice API. Need high-accuracy multilingual with bilingual packs? → Speechmatics speechmatics/standard via WebSocket or Voice API. Need Speechmatics’ broadest language coverage? → Speechmatics speechmatics/linden-1 (preview) via WebSocket or Voice API — vendor-hosted, 67 language codes, no auto-detect (defaults to en). Need real-time transcription with automatic language detection? → Soniox soniox/stt-rt-v4 or soniox/stt-rt-v6 via WebSocket or Voice API. Need term biasing for proper nouns like staff or product names? → Soniox (soniox/stt-rt-v4, soniox/stt-rt-v5, or soniox/stt-rt-v6) in AI Assistants only — pass a context term list, optionally with dynamic variables for caller-specific terms. Not yet available on the plain WebSocket or Voice API surfaces. Need self-hosted multilingual transcription with automatic language detection? → Parakeet nvidia/parakeet-v3 via WebSocket, Voice API, or REST API. Need medical-vocabulary transcription in English? → Parakeet omi-health/omi-med-stt-v1 via WebSocket, Voice API, or REST API. Need self-hosted Arabic or English transcription? → Cohere cohere/ar-stt via WebSocket or Voice API. Requires an explicit ar or en language — no auto-detect. Need turn-based transcription of European languages for voice agents? → Reson8 reson8/turns via WebSocket or Voice API. Need speaker labels on recordings in European languages? → Reson8 reson8/prerecorded via REST API. Need Arabic transcription with Arabic/English code-switching? → Humain humain/realtime via WebSocket or Voice API. Set language to codeswitch or auto.

Specifying the Engine and Model

WebSocket — set via query parameters:
REST API — set via the model body parameter: