Skip to main content

Deepgram Self-Hosted October 2026 Release (261020)

By Steven Van ·

Deepgram's 261020 self-hosted images add FIPS 140-3 variants and a new Flux runtime that no longer loads old .dg model files.

Deepgram has published its self-hosted October 2026 release (261020), with new container images for all four self-hosted services and FIPS 140-3 variants of each.

Container images

  • API: quay.io/deepgram/self-hosted-api:release-261020 (equivalent to 1.192.55)
  • Engine: quay.io/deepgram/self-hosted-engine:release-261020 (equivalent to 3.140.1)
  • License proxy: quay.io/deepgram/self-hosted-license-proxy:release-261020 (equivalent to 1.11.2)
  • Billing: quay.io/deepgram/self-hosted-billing:release-261020 (equivalent to 1.15.1)

The Engine image requires NVIDIA driver version 580 or later, using the open kernel module flavor.

FIPS 140-3 images

Each service also has a FIPS 140-3 image, published under the -fips tag suffix, for example release-261020-fips. These run OpenSSL's FIPS 140-3 validated cryptographic provider. FIPS is enabled by setting [fips] mode = "enabled" in each service's configuration, and only on the -fips images, since the standard images do not include the FIPS provider.

New Flux STT runtime

The release marks an action-required change: Engine now runs Flux speech-to-text on a new runtime that loads only .dgv2 Flux model files, for both flux-general-en and flux-general-multi. The new runtime does not use previously delivered .dg Flux model files. With only a .dg file present and [flux] enabled, Engine exits at startup with "Can't find Flux model".

Force End Turn for Flux

The release note links to the documentation for the Flux ForceEndTurn message. Sending {"type": "ForceEndTurn"} as a WebSocket text frame ends the current turn immediately, for cases such as push-to-talk release, DTMF or IVR events, UI actions or application timeouts. Flux ends the turn on the audio transcribed so far and emits a standard EndOfTurn event with "trigger": "manual". The reported end_of_turn_confidence is the model's real value, not inflated to 1.0.

If no turn is active, the message is ignored and the server returns a FORCE_END_TURN_NO_ACTIVE_TURN warning. It is supported only on the Flux /v2/listen endpoint, and returns an error on /v1/listen (Nova-3).

Deepgram
Deepgram
Power enterprise voice solutions with Deepgram’s Speech-to-Text, Text-to-Speech, and Voice Agent APIs. Real-time, accurate, and built for scale.
View Deepgram →

Original source

Read Deepgram Self-Hosted October 2026 Release (261020) on Creators Toolbox