Google launches Gemini 3.5 Transcribe for intelligent speech-to-text

🚀 RELEASES & LAUNCHES

• Google launches Gemini 3.5 Transcribe for intelligent speech-to-text
  Google launched Gemini 3.5 Transcribe, a speech-to-text model for polished multilingual transcription.

  • Developers can use it in public preview through the Gemini API, Google AI Studio, and Gemini Enterprise Agent Platform.
  • Gemini 3.5 Transcribe supports real-time streaming and recorded-audio transcription, with filler-word removal, custom vocabulary, and over 85 languages.
  • It delivers 4.0% streaming WER and 2.6% non-streaming WER, with timestamps and attribution for up to three speakers.
  • The model powers Rambler on Android and voice features in Gemini for macOS, with Chrome, Docs, Keep, and Gmail integrations coming.
  Sources 📄 Source: blog.google · 📎 Coverage: engadget.com · 👁 via The Verge AI, Google DeepMind Blog

• Ex-Meta scientists launch Perceptron’s vision model for industrial robots
  Perceptron launched Isaac 0.5, an open-weight vision model for industrial robots.

  • Perceptron targets manufacturers, logistics and warehousing, security, mobility, and media companies.
  • Isaac 0.5 helps vision-guided robots perceive, reason, act, and navigate warehouses and factory floors.
  • The model was trained on 1 million hours of general video plus ego and UMI video, alongside petabyte-scale multimodal datasets.
  • Its open-weight release lets companies inspect the model’s parameters and training materials for industrial deployments.
  Sources 📎 Coverage: techcrunch.com · 👁 via TechCrunch AI

• Particle launches Radar to make 130,000+ podcasts searchable by AI agents
  Particle launched Radar, a podcast intelligence platform with web search and API access for AI agents.

  • Podcast listeners, web users, and AI developers can access podcast conversations through Radar.
  • Radar transcribes and analyzes more than 130,000 podcasts, extracting key quotes and highlights.
  • Radar makes podcast content searchable on the web and available to AI agents through an API and MCP.
  Sources 📎 Coverage: techcrunch.com · 👁 via TechCrunch AI

🔬 RESEARCH

• Google introduces GlucoFM foundation model for glucose monitoring
  Google introduced GlucoFM, a lightweight foundation model for continuous glucose monitoring.

  • For researchers and developers building metabolic-health predictions from continuous glucose-monitoring data.
  • Google pre-trained GlucoFM on 109,066 hours of unlabeled CGM data from 477 participant/session records.
  • Its dual-stream architecture separates slower glucose trends from short-term deviations on a 24-hour, five-minute grid while preserving missingness.
  • Across 14 cohort-task evaluations, GlucoFM’s average PR-AUC was 5.8 percentage points higher than the best GluFormer variant; it also achieved the lowest postprandial-response MAE across Dexcom and Libre devices.
  • The model transferred across datasets and adapted with very limited labeled data for diabetes risk, insulin resistance, beta-cell dysfunction, and other metabolic predictions.
  Sources 📎 Coverage: research.google · 👁 via Google Research Blog

• 📄 Source for Z.ai identifies Ox Alpha as GLM-5.3-Flash and releases its weights — z.ai