{"company":{"name":"Deepgram","slug":"deepgram","website":"https://deepgram.com","category":"ai-ml"},"question":"What has Deepgram shipped recently?","answer":"In the last 30 days, Deepgram shipped 31 tracked updates. The most recent was \"Deepgram upgrades Nova-3 models for German and Hindi\" on 2026-11-17.","window":{"days":30,"updateCount":31,"returned":31},"generatedAt":"2026-11-19T05:51:34.915Z","updates":[{"title":"Deepgram upgrades Nova-3 models for German and Hindi","summary":"Deepgram released improved monolingual Nova-3 models for German (de, de-DE, de-AT) and Hindi (hi), improving transcription quality for both batch and streaming workloads. The updates are live for all users without any API request changes, while self-hosted customers must request the updated models from their Deepgram account representative.","date":"2026-11-17","dateIsEstimated":false,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/changelog/2026/11/17","publisher":"developers.deepgram.com"},{"title":"Deepgram adds 36 languages to Nova-3 speech model","summary":"Deepgram added 36 monolingual languages to its Nova-3 model, including Albanian, Amharic, Basque, Icelandic, Swahili, Welsh, Yoruba, and Zulu, available for both batch and streaming workloads. Developers access them by setting model to nova-3 and passing the language code. Self-hosted customers must request the updated models from their Deepgram account representative.","date":"2026-11-16","dateIsEstimated":false,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://developers.deepgram.com/changelog/2026/11/16","publisher":"developers.deepgram.com"},{"title":"Deepgram CLI v0.5.0 changes speak output, init confirmation and telemetry","summary":"Deepgram CLI v0.5.0, released 13 November, adds a --out-file flag to dg speak that refuses to overwrite files without --force, and requires --yes, --install, or --start flags for dg init outside a terminal. dg listen now prints raw server responses for JSON and YAML output and adds a text mode. The release also adds a confirmation guardrail to dg login, structured error fields for JSON output, and expanded telemetry opt-outs. By default, dg has no telemetry destination and sends nothing.","date":"2026-11-13","dateIsEstimated":false,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/changelog/2026/11/13","publisher":"developers.deepgram.com"},{"title":"Deepgram improves Nova-3 models for 13 languages including English variants","summary":"Deepgram released enhanced Nova-3 monolingual models for 13 languages, including new variants for English (Australia/New Zealand), Arabic, Armenian, Bengali, Bulgarian, Danish, Greek, Gujarati, Hebrew, Nepali, Slovak, Tagalog, and Urdu. Improvements apply to both batch and streaming workloads without requiring API changes.","date":"2026-11-12","dateIsEstimated":false,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/changelog/2026/11/12","publisher":"developers.deepgram.com"},{"title":"Build Your First Phone Voice Agent — Deepgram × Vapi Workshop","summary":"Deepgram and Vapi are hosting a hands-on workshop to build a phone voice agent, requiring no prior voice AI experience. Attendees will receive step-by-step guidance, developer support, and dinner while creating a functional agent. The event targets beginners and business owners alike.","date":"2026-11-11","dateIsEstimated":true,"signalType":"event","signalTypeLabel":"Event","sourceUrl":"https://luma.com/deepgram-2jm5?utm_source=direct&utm_medium=referral&utm_campaign=build-first-phone-voice-agent-vapi-workshop&utm_content=deepgram-events-page","publisher":"luma.com"},{"title":"Deepgram Nova-3 adds Basque and Galician to meet Spain’s Ley 10/2025 voice AI requirements","summary":"Deepgram expanded Nova-3 to support Basque and Galician alongside Catalan for voice AI transcription in Spain, aligning with Ley 10/2025’s mandate for co-official language support by December 28, 2026. The update covers EU and self-hosted deployments, enabling real-time transcription for contact centers serving Spanish regions with these languages.","date":"2026-11-10","dateIsEstimated":false,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://deepgram.com/blog/ley-10-2025-voice-ai-readiness","publisher":"deepgram.com"},{"title":"Deepgram Self-Hosted 261110: Nova-3/Flux language hints rejected, callback auth changes, FIPS 140-3 images","summary":"Deepgram’s November 2026 Self-Hosted release (261110) enforces stricter language hint validation for Nova-3 Multilingual and Flux STT, deprecates the `dg-token` callback header in favor of Basic Auth or metadata secrets, and introduces FIPS 140-3 compliant container images. It also optimizes GPU memory for preloaded models and improves Flux TTS shutdown and Portuguese formatting.","date":"2026-11-10","dateIsEstimated":false,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/changelog/2026/11/10","publisher":"developers.deepgram.com"},{"title":"Deepgram powers UpdateAI’s real-time action item detection for Zoom meetings","summary":"Deepgram’s high-accuracy, real-time speech recognition underpins UpdateAI’s AI engine that detects and highlights action items from Zoom meetings for customer success teams. UpdateAI chose Deepgram after testing six providers, citing its superior speed, accuracy, and team responsiveness.","date":"2026-11-09","dateIsEstimated":true,"signalType":"case_study","signalTypeLabel":"Case Study","sourceUrl":"https://deepgram.com/customers/updateai","publisher":"deepgram.com"},{"title":"Telnyx embeds Deepgram Flux for ultra-low-latency voice AI at carrier scale","summary":"Telnyx integrated Deepgram Flux into its global media plane, running the speech-to-text engine on Telnyx-owned GPUs at Points of Presence worldwide. This enables sub-second latency, natural turn-taking, and stable performance under load for real-time voice AI experiences.","date":"2026-11-09","dateIsEstimated":true,"signalType":"integration","signalTypeLabel":"Integration","sourceUrl":"https://deepgram.com/customers/telnyx","publisher":"deepgram.com"},{"title":"Web Summit Kick Off with Vapi and Deepgram","summary":"A networking event in Lisbon co-hosted by Vapi and Deepgram to discuss voice AI trends, production deployment, and the next wave of startups. The event includes a short session on how Vapi and Deepgram help startups build voice agents.","date":"2026-11-09","dateIsEstimated":true,"signalType":"event","signalTypeLabel":"Event","sourceUrl":"https://luma.com/vapi-t98x","publisher":"luma.com"},{"title":"Why we built Scribe","summary":"Deepgram introduced Scribe, a real-time meeting assistant that transcribes audio using Nova-3, extracts action items, and generates structured notes with timestamps and speaker labels. The tool integrates with Notion and Confluence and will be open-sourced in Q2 2026.","date":"2026-11-09","dateIsEstimated":true,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://labs.deepgram.com/posts/why-we-built-scribe","publisher":"labs.deepgram.com"},{"title":"TTS Studio preview","summary":"Deepgram unveiled TTS Studio, a creative workspace for designing voice experiences with real-time previews, voice profiles, and an SSML visual editor. The tool leverages Deepgram’s Aura model for instant streaming output and includes A/B comparison and context window features to refine TTS quality.","date":"2026-11-09","dateIsEstimated":true,"signalType":"product_launch","signalTypeLabel":"Launch","sourceUrl":"https://labs.deepgram.com/posts/tts-studio-preview","publisher":"labs.deepgram.com"},{"title":"Office of the future","summary":"Deepgram Labs ran a month-long internal experiment giving teams custom voice agents for tasks like meeting prep, code deployments, and customer call summaries. Successful use cases included voice-triggered CI/CD and automated meeting briefings, while ambient listening and multi-step tasks proved challenging. The company plans to open-source agent configurations.","date":"2026-11-09","dateIsEstimated":true,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://labs.deepgram.com/posts/office-of-the-future","publisher":"labs.deepgram.com"},{"title":"Claude Code for the win","summary":"Deepgram’s engineering team replaced manual coding with Claude Code, cutting prototyping time from days to hours and enforcing consistency across their monorepo. The tool improved documentation quality and reduced repetitive tasks, though they emphasize human oversight for critical decisions.","date":"2026-11-09","dateIsEstimated":true,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://labs.deepgram.com/posts/claude-code-for-the-win","publisher":"labs.deepgram.com"},{"title":"Shipping AI Works v0.1","summary":"Deepgram released AI Works v0.1, a visual workspace for building and deploying AI pipelines with one click. The tool uses React Flow for a canvas-based editor, integrates Deepgram’s speech APIs for voice-triggered workflows, and runs on a lightweight TypeScript execution engine. The initial version includes six core node types: input, output, LLM, transform, condition, and API call.","date":"2026-11-09","dateIsEstimated":true,"signalType":"product_launch","signalTypeLabel":"Launch","sourceUrl":"https://labs.deepgram.com/posts/shipping-ai-works-v01","publisher":"labs.deepgram.com"},{"title":"Deepgram launches Marketplace with 19 vetted voice AI partners","summary":"Deepgram introduced its Marketplace to simplify building voice AI stacks by aggregating vetted third-party partners across 8 categories, including models, testing, audio enhancement, and telephony. Eligible customers can use shared credits to pay for partner products under one contract, reducing vendor overhead.","date":"2026-11-04","dateIsEstimated":false,"signalType":"integration","signalTypeLabel":"Integration","sourceUrl":"https://deepgram.com/blog/introducing-deepgram-marketplace","publisher":"deepgram.com"},{"title":"Deepgram adds Extra Metadata feature to API responses","summary":"Deepgram introduced Extra Metadata, allowing users to attach arbitrary key-value pairs to API requests that are returned in responses for downstream processing. The feature supports up to 2048 characters per pair and can be enabled via the `extra` query parameter. It complements existing Tagging functionality but is designed for data passing rather than usage tracking.","date":"2026-10-30","dateIsEstimated":true,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/docs/extra-metadata","publisher":"developers.deepgram.com"},{"title":"Deepgram launches MCP servers and skills marketplace for AI coding tools","summary":"Deepgram introduced a CLI with a built-in MCP server, a docs MCP server for querying documentation, and a skills marketplace with structured reference material for AI-powered developer tools. These tools enable direct API access and documentation queries from coding assistants like Claude Code and Cursor.","date":"2026-10-29","dateIsEstimated":true,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/developer-tools/agentic-tools","publisher":"developers.deepgram.com"},{"title":"Deepgram launches streaming text-to-speech WebSocket API with SDKs","summary":"Deepgram introduced a new WebSocket-based text-to-speech (TTS) API that converts streaming text into speech in real time. The feature supports multiple SDKs (JavaScript, Python, Go, C#, Java) and uses the 'aura-2-thalia-en' model, with detailed code examples for implementation.","date":"2026-10-29","dateIsEstimated":true,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://developers.deepgram.com/docs/streaming-text-to-speech","publisher":"developers.deepgram.com"},{"title":"Deepgram expands Aura-2 TTS voices with 30+ English options and broader language support","summary":"Deepgram’s Aura-2 text-to-speech API now offers over 30 English voices across American, British, Australian, Irish, and Filipino accents, alongside expanded language support including Spanish, German, French, Dutch, Italian, and Japanese. The update introduces new voices with distinct characteristics for varied use cases like customer service, IVR, advertising, and storytelling.","date":"2026-10-29","dateIsEstimated":true,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/docs/tts-models","publisher":"developers.deepgram.com"},{"title":"Deepgram launches Think-Fast and phoneLLM mini LLMs for voice agents","summary":"Deepgram introduced two new LLMs—Think-Fast for instruction-following voice conversations and phoneLLM mini for speed-optimized pipelines—both available during Early Access. Think-Fast supports cascaded voice agents with tool calling, while phoneLLM mini is a 30B-parameter MoE model (3.5B active) from Pipecat, prioritizing low latency.","date":"2026-10-29","dateIsEstimated":true,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://developers.deepgram.com/docs/llm/models","publisher":"developers.deepgram.com"},{"title":"Deepgram launches Early Access LLM endpoint /v1/think with Think-Fast model","summary":"Deepgram introduced a new LLM endpoint (/v1/think) and model (think-fast) for Early Access customers, available only to contract holders. The endpoint supports stateless chat completions with system/user messages, returns usage metrics, and lacks SDK support during Early Access. Production use is discouraged.","date":"2026-10-29","dateIsEstimated":true,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://developers.deepgram.com/docs/llm/getting-started","publisher":"developers.deepgram.com"},{"title":"Deepgram launches Early Access LLMs for voice workloads with Think-Fast and phoneLLM mini","summary":"Deepgram introduced two Early Access LLMs—Think-Fast and phoneLLM mini—designed for voice agent pipelines. Think-Fast supports cascaded agent workflows with tool calling, while phoneLLM mini prioritizes speed and size. Both models are accessible via a new endpoint and require enrollment for contract customers.","date":"2026-10-29","dateIsEstimated":true,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/docs/llm","publisher":"developers.deepgram.com"},{"title":"Deepgram launches self-hosted voice AI for cloud and on-premises deployment","summary":"Deepgram introduced a self-hosted version of its voice AI platform, enabling deployment on customer infrastructure for enhanced performance and privacy. The offering supports real-time inference with <200ms latency, data residency compliance, and seamless integration with Kubernetes, Docker, and other orchestrators. Enterprise customers can leverage AWS/GCP marketplace agreements for cost savings.","date":"2026-10-29","dateIsEstimated":true,"signalType":"product_launch","signalTypeLabel":"Launch","sourceUrl":"https://deepgram.com/self-hosted","publisher":"deepgram.com"},{"title":"Deepgram launches Voice Agent Platform with Think-Live/Think-Fast LLMs and Real-Time Context Standard","summary":"Deepgram introduced the Real-Time Context Standard (RCS) for voice agents, two specialized LLMs (Think-Live for conversation management and Think-Fast for task execution), and a new Voice Agent Platform built on the open-source Pipecat framework. The platform aims to enable agents to handle interruptions, corrections, and multilingual interactions seamlessly.","date":"2026-10-29","dateIsEstimated":false,"signalType":"product_launch","signalTypeLabel":"Launch","sourceUrl":"https://deepgram.com/blog/voice-agents-that-keep-the-thread","publisher":"deepgram.com"},{"title":"Deepgram acquires Daily and Pipecat to launch first open, production-ready voice AI stack","summary":"Deepgram announced the acquisition of Daily, creators of the open-source Pipecat voice agent framework, to integrate Pipecat’s orchestration layer with Deepgram’s speech models and Daily’s global real-time infrastructure. The combined platform aims to provide the first complete, open, and production-grade stack for conversational voice AI, enabling developers to build scalable agents while maintaining model and architecture choice. Pipecat will remain fully open source under its BSD V2 license, and a new Pipecat Foundation will ensure vendor neutrality.","date":"2026-10-29","dateIsEstimated":false,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://deepgram.com/blog/daily-pipecat-deepgram","publisher":"deepgram.com"},{"title":"Deepgram deprecates dg-token for callback authentication","summary":"Deepgram is retiring the `dg-token` header for verifying callback requests across its Speech-to-Speech, Text-to-Speech, and Text Intelligence APIs. Developers must migrate to Basic Auth or use a secret passed via Extra Metadata for Speech-to-Text callbacks. The change affects all callback integrations and requires immediate updates to avoid failed verifications.","date":"2026-10-27","dateIsEstimated":false,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/changelog/2026/10/27","publisher":"developers.deepgram.com"},{"title":"Deepgram Nova-3 adds multilingual transcription with language hints","summary":"Deepgram’s Nova-3 model now supports multilingual transcription for 55 languages, including mid-conversation language switching, via the `language=multi` parameter. Users can optionally bias detection with the `language_hint` parameter for known languages or sets, while auto-detection remains available for unknown inputs.","date":"2026-10-23","dateIsEstimated":true,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://developers.deepgram.com/docs/nova-3-language-prompting.md","publisher":"developers.deepgram.com"},{"title":"Deepgram Nova-3 multilingual model expands to 55 languages with language hinting","summary":"Deepgram’s Nova-3 multilingual model now supports 55 languages (up from 10) and introduces language hinting to bias detection toward expected languages. Code-switching and per-word language tags remain supported on the same `/v1/listen` endpoints for both pre-recorded and streaming audio.","date":"2026-10-23","dateIsEstimated":false,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/changelog/2026/10/23.md","publisher":"developers.deepgram.com"},{"title":"Resources Hub","summary":"Deepgram announced the acquisition of Daily, the company behind Pipecat, to create the first complete, production-ready open stack for conversational voice AI. This move integrates Daily's real-time communication infrastructure with Deepgram's speech technologies.","date":"2026-10-23","dateIsEstimated":false,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://deepgram.com/resources","publisher":"deepgram.com"},{"title":"Deepgram launches EU, AU, and India regional endpoints for data residency","summary":"Deepgram introduced regional endpoints in the EU (`api.eu.deepgram.com`), Australia (`api.au.deepgram.com`), and India (`api.in.deepgram.com`) to ensure data processing occurs within the specified geography. Customers can switch to these endpoints by replacing the base URL in their SDKs or API requests while using existing credentials. The EU endpoint excludes Whisper models, while AU and India endpoints mirror this limitation.","date":"2026-10-21","dateIsEstimated":true,"signalType":"feature_update","signalTypeLabel":"Feature","sourceUrl":"https://developers.deepgram.com/reference/regional-endpoints.md","publisher":"developers.deepgram.com"}],"attribution":{"source":"Spyingbee","url":"https://spyingbee.com/updates/deepgram","citation":"Spyingbee, \"Deepgram updates\", https://spyingbee.com/updates/deepgram (retrieved 2026-11-19)"}}