Flux Feature Overview | Deepgram's Docs

Model Selection

Feature	Language(s)
Model	`flux-general-en` (English), `flux-general-multi` (10 languages — Language Prompting)
Version	All available

Language Configuration

Feature	Model	Description
Language Prompting	`flux-general-multi` only	Bias output toward specific languages with `language_hint`
Language Detection	`flux-general-multi` only	`languages` field on TurnInfo events reports detected languages

Formatting

Feature	Language(s)
Profanity Filter	All available
Numerals	English + Multilingual (excludes Hindi & Japanese)
Redaction	Number redaction only (`numbers`, `aggressive_numbers`) — English + Multilingual

Transcription

Feature	Language(s)
Word-level Timestamps	All available

Custom Vocabulary

Feature	Language(s)
Keyterm Prompting	All available

Media Input Settings

Feature	Language(s)
Sample rate	All available
Encoding	All available

End-of-Turn Configuration

Feature	Description
Configurable Parameters	Tune `eot_threshold`, `eager_eot_threshold`, and `eot_timeout_ms` for optimal turn detection

Flux’s end-of-turn detection is configurable to match your use case. Use eot_threshold for standard turn detection, or enable eager_eot_threshold for ultra-low latency response generation. See the End-of-Turn Configuration for detailed configuration guidance.

Control Messages

Feature	Description
Configure	Update keyterms and thresholds mid-stream
Close Stream	Close the WebSocket stream
Force End Turn	End the current turn immediately from an external signal

Rate Limits

For information on Deepgram’s Concurrency Rate Limits, refer to our API Rate Limits Documentation.

Deepgram Self-Hosted

Flux is now available for self-hosted deployments!

For setup instructions, see Using the Flux Model. For more information about Deepgram’s self-hosted solution, visit our Self-Hosted Introduction.