Tencent Open-Sources Hy3 Model, TECNO Teams With Google AI Plus for Free Cloud Storage - AI Daily Brief (Jul 7, 06:00)
Audio in Mandarin Chinese · English transcript below
⚡ Tencent's Hy3 surpasses GLM open-source; Reddit deploys AI anti-spam; Siri gains speed control; AMD kits hit premium pricing; nations race AI factories; quantization revolution accelerates...
AI is increasingly locked in a "last mile" race centered on end-user experience. Transsion has partnered with Google to bundle AI cloud storage for emerging markets, while Apple is defending its premium positioning with customizable Siri voices in iOS 27—both converging on the same imperative: making AI genuinely embedded in daily life. What deserves particular attention is Tencent's move to open-source its Hy3 model under the Apache license, targeting foundational infrastructure. This bifurcation—competing on experience at the top layer while contesting standards at the bottom—has become the most defining structural pattern in today's AI landscape.
Today's Top 3 Headlines
- AI Model Release
🚀 Tencent open-sources Hy3 under Apache 2.0, beats GLM-5.1 in blind tests, lags in coding
Tencent Hunyuan releases Hy3 (295B total / 21B active params, Apache 2.0), scoring 2.67 in blind tests vs GLM-5.1's 2.51, though coding lags GLM-5.2 on SWE-bench. For devs, Hy3 excels at search and tool orchestration tasks, offering a strong open-weight alternative for Agent workflows.
Source ↗ - AI Industry News
🤖 TECNO partners with Google AI Plus: free 2TB cloud storage, saving 2,000 rupees/month
Transsion partners with Google AI Plus to offer 3 months free 2TB cloud storage for Pakistan SPARK 50 users from Dec 2025, bundling Gemini AI, AI photo editing and Ask Photos, saving ~PKR 2,000/month. For consumers, this bundles premium AI services into affordable devices; for Google, it locks emerging-market users into its AI ecosystem before rivals like Doubao or Qwen expand globally.
Source ↗ - Technology
🤖 Reddit Blocks 23M AI Spam Views Daily with LLM Tools
Reddit revealed its LLM tools now block 23M AI spam views daily, catch ~25K new posts, and cut spam exposure by 20%. For platform operators, this signals LLM-driven spam defense will increasingly rely on AI-vs-AI combat, requiring detection models to iterate continuously.
Source ↗
+6 more headlines
- 🤖 iOS 27 Beta 3: Siri Gains Custom Speech Rate & Expressiveness
- 🤖 AMD Launches $4,000 AI Dev Kit, Same Hardware as Previous Gen
- 🤖 Global AI Race: 5 Drivers Power Localized LLM Deployment
- 🤖 UTA Co-Head: How AI Era Helps Human Creators
- 🤖 Structured Output & Feedback Loops Before RAG Answer Delivery
- 🤖 Quantization Over Compute: GLM 5.2, Qwen & DeepSeek Reshape AI Revolution via Model Compression