5 stories that day · 3 models · 2 open source
Published daily at 7 PM ET · Updated · RSS
ModelTop story
MAI-Transcribe-2-Streaming turns live speech into text in 60 languages, and MAI-Voice-2.1 reads text aloud in 23. Microsoft says both are available now in Microsoft Foundry.
Microsoft · Original source (opens in a new tab)
Model
Griffin watches, listens and replies with a talking face in a live video call. Tavus says 48% of people thought it was a real person after one minute; only selected testers can try it for now.
Open source
Strands Decider, a small open model that decides instead of writing (opens the original source in a new tab)A fast model for AI agents: rather than writing text, it picks between options or gives a score, with a confidence attached to every answer. The code is open source under the Apache-2.0 licence.
Strands Agents · Original source (opens in a new tab)
Model
Share your phone camera and talk: Gemini describes what it sees, like the small print on food labels. Google is launching it on compatible Android phones for people who are blind or have low vision.
Google · Original source (opens in a new tab)
Open source
Ai2 releases Olmo-core 3 for training giant mixture-of-experts models (opens the original source in a new tab)Open software for training mixture-of-experts models, which send each input to only a few specialist parts. Ai2 says it is built to scale that training into the trillion-parameter range.
Notable stories from the past few days. Each shows its own date.
