Tencent Puts 33-Language Translation on Devices Offline
Tencent’s new Hy Translation app brings Hy-MT2 to phones with offline text, voice and image translation across 33 languages and Chinese dialects.
September 23, 2026
All briefings and quick hits published that day.
- ⚡ Quick hits
- Safety OpenAI introduced a tiered framework and case studies for reporting model misalignment. InfoQ 中文 ↗
- AIGC Meshy added Auto Split, allowing users to describe which parts an AI-generated 3D asset should separate. X / MeshyAI ↗
- Industry A new embodied-intelligence theme park opened with more than 300 robots for visitors. IT之家 ↗
- Industry General Motors reportedly compares employees’ AI usage rates in monthly internal competitions. IT之家 ↗
- Policy AI dominated discussions at Climate Week, reflecting growing expectations for its role in climate action. MIT Technology Review ↗
- Industry Kaivid Labs documented migrating more than 400,000 document vectors from Milvus to Qdrant. X / Qdrant ↗
- ⚡ Quick hits
- AIGC Dreamina promoted its web app and announced winners from a new creative challenge. X / Dreamina AI ↗
- Agents Contrastive-LM released CLM-8B, an open system-one model that reportedly scores agent actions up to nine times faster than Jev. MarkTechPost ↗
- Safety Transluce reported early rogue-agent activity and hacking attempts discovered through urlquery.net. Transluce ↗
- Open Source Tsinghua University and Wuwen Xinqiong open-sourced RLark, a cloud-native platform for managing embodied-AI robots and cross-cluster tasks. 量子位 ↗
- Industry NVIDIA said more doctors will receive free access to OpenEvidence through an initiative involving Anthropic. X / NVIDIAAI ↗
- Research Tencent Hunyuan published new research on improving reinforcement-learning efficiency across larger GPU clusters and datasets. X / TencentHunyuan ↗
- AIGC Alibaba demonstrated a video workflow generating cinematic motion, 1080p output, and sound from one prompt. X / Alibaba Wan ↗
- Agents Amazon introduced a plugin that lets sellers manage their entire e-commerce business with AI. SiliconANGLE ↗
- Industry Ollama began offering pay-as-you-go credits for cloud models without requiring a subscription. X / Ollama ↗
Meta Expands Muse Agent Across Glasses, Apps and Services
Meta is extending Muse from a standalone personal agent into a cross-device assistant with new connectors, computer control and wearable access.
Cursor adds deployment monitoring and security review bots
Cursor launched Rollouts and Security Review, two enterprise bots that monitor production changes and inspect pull requests for exploitable flaws.
- ⚡ Quick hits
- Funding Basecamp Research raised $140 million to use evolutionary data and AI for drug discovery. SiliconANGLE ↗
- Funding Enveda raised $311 million to advance nature-derived AI drug candidates into clinical trials. TechCrunch ↗
- Safety OpenAI introduced MentalHealthBench, an open benchmark designed to evaluate realistic conversations across the full spectrum of mental-health needs. X / OpenAI ↗
Claude identifies a CRISPR-like enzyme system in phages
Anthropic says Claude agents found an uncharacterized reverse-transcriptase system after screening 200,000 proteins and guiding lab validation.
- ⚡ Quick hits
- Open Source NVIDIA released Nemotron 3 Diarization, a 100-million-parameter open-weight model that tracks up to eight overlapping speakers in real time. MarkTechPost ↗
- Agents Cognition added native Microsoft Teams support and connected Devin to Outlook, Calendar, OneDrive, SharePoint, and Directory. X / Cognition ↗
- Industry Google expanded Beam, its glasses-free 3D video communication platform, with new regions, partners, and customers. Google Blog ↗
- Research Microsoft Research found that offloading robot inference to edge or cloud GPUs improved task success, responsiveness, battery life, and model capacity. Microsoft Research ↗
- Industry OpenAI said its Academy has reached more than four million people and will expand through a community trainer program. OpenAI ↗
- Models An Anthropic engineer said Claude’s writing quality declined because training increasingly optimized for AI-readable prose rather than human readers. IT之家 ↗
Google launches Gemini models for programmable voice design
Google’s new Gemini 3.8 Flash TTS models let users design voices, direct performances line by line and generate speech across 100 languages.
- ⚡ Quick hits
- AIGC Google’s Gemini 3.8 Flash TTS and Flash-Lite TTS let users design voices from text prompts across more than 100 languages. Google DeepMind ↗
- Agents Meta’s Muse AI agent reportedly attracted 500,000 users in one week, while critics alleged it copied OpenClaw. The Decoder ↗
- AIGC Recraft launched V4.1 Flash, claiming image generation in 1.3 seconds and availability through official partner fal. X / Recraft ↗
- AIGC YouTube announced AI tools for Creator Studio, including script coaching, smart thumbnails, Gemini editing, and customizable recommendation feeds. The Verge ↗
- AIGC ElevenLabs released Scribe v2 Medical, offering HIPAA-eligible deployment with zero-retention mode and pricing starting at $0.22 per hour. X / ElevenLabs ↗
- Funding Ema raised $77 million to deploy AI employees across enterprise human resources, IT, and finance departments. SiliconANGLE ↗
Alibaba introduces Qwen Intelligence for mobile agents
Alibaba unveiled Qwen Intelligence with mobile planning, device-use and creative agents, aiming to make smartphone assistants act across apps.