Deepgram Acquisition: Pipecat Maker Daily And The New Voice AI Infrastructure Race
The Deepgram acquisition of Daily, maker of the open source Pipecat framework, unites voice models with real-time infrastructure. Here is what changes for developers.
The Deepgram acquisition of Daily, maker of the open source Pipecat framework, unites voice models with real-time infrastructure. Here is what changes for developers.
Nvidia open transcription is now live: Nemotron 3.5 ASR Streaming 0.6B transcribes 40 language-locales in real time, sustains up to 240 concurrent streams per H100 at 80ms latency, and ships under an OpenMDW-1.1 license for commercial use. We unpack the cache-aware FastConformer-RNNT architecture, the three language tiers, the Granary-v2 data push behind it, how it compares with Parakeet, Canary and Whisper, and the pilots voice teams should run first.
A hidden Voice tab with a Preview badge has been spotted in Claude’s web sidebar, beside the existing voice mode rather than replacing it. We explain what the screenshot shows, how it ties to Anthropic’s new voice data opt-in, what an upgraded voice feature could be, and how Claude compares with ChatGPT GPT-Live, Gemini 3.8 Live and Grok Voice.
Onepin, from speech-evaluation company Podonos, is an AI voice production agent that routes each line of a script to a text-to-speech engine, scores it for naturalness, clarity and pronunciation, and fixes what fails, in 10 languages. We explain the pipeline, the pricing per validated hour, where Onepin’s own claims disagree, who it suits and what UK teams should check.
ChatGPT Pro Max, a $500-a-month ChatGPT plan, appeared in OpenAI’s web app code and in a Codex commit on 24 and 25 September 2026. The same week Google released Gemini 3.8 Live and made Live Avatar generally available. This article covers the leak evidence, the renamed Pro ladder, the Gemini benchmarks and pricing, and what businesses should do.
Google is rolling out real-time voice conversations in Gemini Notebook, starting with Google AI Ultra subscribers on mobile. We set out what the feature does, the four restrictions buried in the announcement, the Gemini 3.8 Live model it almost certainly runs on, the four other study features shipped alongside it, and whether the feature justifies an Ultra subscription.
TestingCatalog spotted Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on Google Cloud’s console quotas page on 15 September 2026. We set out what a quota entry proves, where the model would sit in Google’s Live API lineup, how it compares with OpenAI’s GPT-Live-1, and what Musk’s posts and three prediction markets say about Grok 4.7.
Agora Meeting Copilot, an MIT-licensed demo, moved its AI teammate to OpenAI’s GPT-Live-1 on 11 September 2026. We read its 13,152 lines: a wake-phrase gate added after a failed demo, a generally available model sent through an Agora route labelled alpha preview, and $0.15 a minute in Agora and OpenAI charges while the AI stays silent.
OpenAI launched GPT-Live-1 in the API and the Agents API in public beta on 10 September 2026. We read both announcements against 21 pages of OpenAI’s documentation. Four of GPT-Live-1’s seven benchmark charts score it with a GPT-6 Astra or Terra backend attached, its $0.05-a-minute rate bills silence, and the two products sit in different data-control categories: GPT-Live-1 is eligible for Zero Data Retention and European processing, the Agents API is not.
Gemini 3.5 Transcribe is now listed on AIxploria, the AI tools directory that catalogues thousands of AI sites across more than fifty categories. The new card files Google’s speech-to-text model under Transcriber, where it already sits at number eight, prices it “Freemium”, rates it 4.5 out of 5 and stamps it with a Verified Tool […]