Releasesby GlyphImproved Models Released for Multiple Languages
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.
Also shipped Aug 4, 2026Deepgram in August 2026 →
Workspace sharing in Declarative Native Apps (Public Preview)
Workspace sharing is now available in public preview for Declarative Native Apps. Providers can share a whole directory of files and folders by declaring a workspace in the application package manifest. Consumers get a read-only workspace when they install the app, and can browse the folder structure and open the files.
Gateway Monitoring & A/B Testing (General availability)
Gateway Monitoring & A/B Testing is now generally available. Use gateway model monitors to track drift and performance for inference services behind a Snowflake Gateway, compare baseline and challenger services during live traffic splits, and evaluate results in Snowsight or with SQL. For more information, see Gateway monitoring & A/B testing .
Multi-party Approval (General availability)
Multi-party Approval (MPA) is now generally available. When a Multi-party Approval policy is attached to your account, MPA enforces a mandatory second-person (“four-eyes”) verification step for critical account operations. A Multi-party Approval policy defines which operations require approval and who can approve them.
Unity Gateway is now generally available
Unity Gateway is now generally available. Unity Gateway is the Databricks governance solution for enterprise AI, part of Unity Catalog. With Unity Gateway you can: Control which AI services teams can use. Route and manage AI traffic across providers. Govern MCP servers to control access and costs. Monitor usage, cost, access, and lineage from one place.
Sharing Delta tables with Iceberg reads enabled is now generally available
Sharing Delta tables with Iceberg reads enabled is now generally available. See Delta Lake feature support matrix .
Sharing Delta tables with deletion vectors and column mapping is now generally available
Sharing Delta tables with deletion vectors or column mapping enabled is now generally available. See Add tables with deletion vectors or column mapping to a share .
More from DeepgramEvery Deepgram release →
🆕 Improved Models
Batch models Croatian ( hr ) Hungarian ( hu ) Lithuanian ( lt ) Korean ( ko , ko-KR ) Japanese ( ja ) Gujarati ( gu , gu-IN ) Estonian ( et ) Spanish ( es , es-419 ) Marathi ( mr ) Belarusian ( be ) Latvian ( lv ) Streaming models Croatian ( hr ) Lithuanian ( lt ) Korean ( ko , ko-KR ) Japanese ( ja ) Gujarati ( gu , gu-IN ) Estonian ( et ) Spanish ( es ...
Nova-3 Model Update
🌏 Nova-3 monolingual models now support Punjabi and Nepali with the following language codes: Punjabi: pa , pa-IN Nepali: ne Batch and streaming models are both available for these languages. Access this model by setting model="nova-3" and the relevant language code in your request.
Improved Models Released for Multiple Languages
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.
🆕 Improved Models
Batch models Bulgarian ( bg ) Bengali ( bn ) Bosnian ( bs ) Belarusian ( be ) Czech ( cs ) Norwegian ( no ) Streaming models Bulgarian ( bg ) Bosnian ( bs ) Czech ( cs ) Norwegian ( no ) Latvian ( lv ) You do not need to change your API requests to access these improvements; the updates are live for all users.
UpdateListen: switch model and language mid-session
UpdateListen now accepts model and language changes, so you can switch speech-to-text models and languages without restarting the session. Previously the provider identity had to match the current session.
STT latency in LatencyReport
LatencyReport now includes stt_latency , the time from audio received to transcript produced. Every turn reports the full STT → LLM → TTS breakdown. { "type": "LatencyReport", "stt_latency": 0.12, "ttt_token_latency": 0.34, "tts_latency": 0.18, "total_latency": 0.64 } For details, see the Latency Report documentation .
Maximum session length of 2 hours
Voice Agent sessions now close automatically after 2 hours. Two new codes mark the limit: Time into the session Event Code 1 hour 55 minutes Warning MAXIMUM_SESSION_LENGTH_APPROACHING 2 hours Error (terminal) MAXIMUM_SESSION_LENGTH_REACHED KeepAlive does not extend the limit.
Google Enterprise Agent API support
Managed Google LLMs now run on either of Google's APIs: the AI Studio API or the Gemini Enterprise Agent (GEA) API, formerly the Vertex API. GEA adds regional Google endpoints for models where available. Select the API with think.provider.version from any Deepgram Voice Agent endpoint: ai-studio-v1beta for AI Studio, gemini-enterprise-agent-v1 for GEA.