Changelog

🆕 Numeral Formatting for Flux

Flux now supports the numerals feature, which converts numbers from written format to numerical format (for example, “nine hundred” becomes “900”). Numerals are available on both Flux models:

• Flux English (model=flux-general-en) — full numeral formatting.
• Flux Multilingual (model=flux-general-multi) — numeral formatting for English, Spanish, French, German, Russian, Portuguese, Italian, and Dutch. Numeral formatting is not currently supported for Hindi or Japanese.

We have released improved Nova-3 monolingual models for the following languages. These updates enhance transcription quality and accuracy for both batch and streaming workloads.

Deepgram Self-Hosted July 2026 Release (260714)

Container Images (release 260714)

• quay.io/deepgram/self-hosted-api:release-260714

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-api:1.192.5-1

• quay.io/deepgram/self-hosted-engine:release-260714

  • Equivalent image to:

    • quay.io/deepgram/self-hosted-engine:3.119.1-3

  • Minimum required NVIDIA driver version: >=570.172.08

• quay.io/deepgram/self-hosted-license-proxy:release-260714

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-license-proxy:1.10.2

• quay.io/deepgram/self-hosted-billing:release-260714

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-billing:1.13.1

InjectAgentMessage: behavior field and new interrupt mode

The InjectAgentMessage message now accepts a behavior field that controls how an injected message interacts with any in-progress user or agent turn. The default and queue behaviors were previously documented; the new interrupt behavior is introduced with this release.

🆕 Word-Level Timestamps Now Available in Flux

Flux now includes word-level timestamps in its responses. Each word in the words array carries start and end times (type double) alongside the existing word and confidence values.

Example response:

1
{
2
  "event": "EndOfTurn",
3
  "turn_index": 0,
4
  "transcript": "Hello, how can I help you?",
5
  "words": [
6
    {

claude-sonnet-5 is now available as a managed Anthropic LLM in the Voice Agent API. This Advanced tier model delivers improved reasoning and conversational quality for your voice agents.

Set the model in your agent configuration:

1
{
2
  "agent": {
3
    "think": {
4
      "provider": {
5
        "type": "anthropic",
6
        "model": "claude-sonnet-5",

Deepgram Self-Hosted June 2026 Release (260630)

Container Images (release 260630)

• quay.io/deepgram/self-hosted-api:release-260630

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-api:1.192.3-1

• quay.io/deepgram/self-hosted-engine:release-260630

  • Equivalent image to:

    • quay.io/deepgram/self-hosted-engine:3.119.1-1

  • Minimum required NVIDIA driver version: >=570.172.08

• quay.io/deepgram/self-hosted-license-proxy:release-260630

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-license-proxy:1.10.1-2

• quay.io/deepgram/self-hosted-billing:release-260630

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-billing:1.13.0

Australia Endpoint Now Generally Available

The Deepgram Australia endpoint (api.au.deepgram.com) is now generally available for customers requiring data processing within Australia.

Supported APIs

The Australia endpoint supports the following Deepgram APIs:

You can now update the Listen configuration during a conversation without restarting the session. Send an UpdateListen message to tune:

• eot_threshold — end-of-turn confidence threshold
• eager_eot_threshold — eager end-of-turn confidence threshold
• eot_timeout_ms — hard timeout for turn completion
• keyterms — keyword boost list
• language_hints — language bias for multilingual models

Deepgram Self-Hosted June 2026 Release (260611)

Container Images (release 260611)

• quay.io/deepgram/self-hosted-api:release-260611

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-api:1.191.0-1

• quay.io/deepgram/self-hosted-engine:release-260611

  • Equivalent image to:

    • quay.io/deepgram/self-hosted-engine:3.118.0-1

  • Minimum required NVIDIA driver version: >=570.172.08

• quay.io/deepgram/self-hosted-license-proxy:release-260611

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-license-proxy:1.10.1-1

• quay.io/deepgram/self-hosted-billing:release-260611

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-billing:1.13.0

🆕 Improved Nova-3 Medical Batch Model Released

We’ve released an upgraded Nova-3 Medical batch model with improved medical term recognition.

Key Improvements:

• Expands the model’s medical vocabulary with new medical terms
• Boosts medical term recognition to 97.20% KRR
• Maintains WER parity with no observed regression on common or conversational terms

Deepgram Self-Hosted May 2026 Release (260528)

Container Images (release 260528)

• quay.io/deepgram/self-hosted-api:release-260528

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-api:1.188.1

• quay.io/deepgram/self-hosted-engine:release-260528

  • Equivalent image to:

    • quay.io/deepgram/self-hosted-engine:3.117.0

  • Minimum required NVIDIA driver version: >=570.172.08

• quay.io/deepgram/self-hosted-license-proxy:release-260528

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-license-proxy:1.10.1

• quay.io/deepgram/self-hosted-billing:release-260528

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-billing:1.13.0

gemini-3.5-flash is now available as a managed Google LLM in the Voice Agent API. This Standard tier model brings improved performance and efficiency to your voice agents.

Set the model in your agent configuration:

1
{
2
  "agent": {
3
    "think": {
4
      "provider": {
5
        "type": "google",
6
        "model": "gemini-3.5-flash",

🆕 Profanity Filtering for Multilingual Models

Deepgram’s Profanity Filtering feature is now available for all multilingual models: Nova-2 multilingual, Nova-3 multilingual, and Flux multilingual (language=multi). You can enable profanity filtering in your API requests by setting the profanity_filter=true parameter. When enabled, inappropriate language is automatically replaced with asterisks (****) in the transcript.

gemini-3.1-flash-lite is now available as a managed Google LLM in the Voice Agent API. This Standard tier model replaces the preview version.

Set the model in your agent configuration:

1
{
2
  "agent": {
3
    "think": {
4
      "provider": {
5
        "type": "google",
6
        "model": "gemini-3.1-flash-lite",

Supported languages and language codes:

• Russian (ru)
• Romanian (ro)
• Hebrew (he)

You can now use Deepgram’s Numerals feature with monolingual models for Russian, Romanian, and Hebrew. Numerals converts spoken numbers into digits (for example, “three hundred” → “300”) in your transcript, helping you create more accurate and easily processed results.

Deepgram Self-Hosted May 2026 Release (260514)

Container Images (release 260514)

• quay.io/deepgram/self-hosted-api:release-260514

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-api:1.187.0

• quay.io/deepgram/self-hosted-engine:release-260514

  • Equivalent image to:

    • quay.io/deepgram/self-hosted-engine:3.117.0

  • Minimum required NVIDIA driver version: >=570.172.08

• quay.io/deepgram/self-hosted-license-proxy:release-260514

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-license-proxy:1.10.1

• quay.io/deepgram/self-hosted-billing:release-260514

  • Equivalent image to:
    • quay.io/deepgram/self-hosted-billing:1.13.0

We’re excited to announce the release of profanity filtering support for over 50 monolingual languages. Deepgram’s profanity filter automatically detects and redacts offensive language in transcripts, helping you produce cleaner and safer content across a wide range of languages.

A new batch diarization model is available today via the diarize_model API parameter.

Deepgram is rolling out v2 of our batch speaker diarization model. v2 is a new architecture available today on an opt-in basis through the new diarize_model parameter. In side-by-side human evaluation, v2 was preferred 3.3× over our current production diarizer (v1), with the largest gains on contact-center audio — median CER reduced roughly 80% compared to the prior version of the diarization model. Customers using diarize=true are unaffected.

A new round of SDK updates is now available across JavaScript, Rust, Python, and Java. This release brings Flux multilingual support to Rust, restores the Agent interface in JavaScript, ships a Python bugfix for WebSocket query parameters, and delivers a breaking Java release with reconnect improvements.