🚀 Gemini Omni 1.2 Flash&Supermicro/IBM/NVIDIA Shared KV|AI Daily Brief
Audio in Mandarin Chinese · English transcript below
⚡ Gemini Omni 1.2 Flash; 56x inference; 280B ERNIE open-source: AI race hits cost.
Foundation-model iteration and vertical deployment are accelerating in tandem, forming the central storyline of today’s AI competition. Google DeepMind has sharply reduced video-generation costs while extending clip length, and Baidu has set new benchmarks with a 280-billion-parameter open-source model; together, they are pushing the boundaries of multimodal capabilities and the open-source ecosystem. At the same time, OpenAI’s release of GPT-6 Cyber, dedicated to cybersecurity, shows that leading vendors are rapidly converting general-purpose capabilities into commercial moats in specialized domains.
Today's Top 3 Headlines
- AI Industry News
🤖 Gemini Omni 1.2 Flash launches: 15-second video, 30% price cut
Google DeepMind's Gemini Omni 1.2 Flash generates 3–15s video and cuts pricing 30% to $0.07/sec at 720p. It enables cheaper scalable branded video and consistent characters/colors.
Source ↗ - Technology
🤖 Supermicro, IBM, NVIDIA launch shared KV cache, up to 56x faster first token
Supermicro/IBM/NVIDIA offer shared KV-cache storage for reusable inference context off GPU memory;tests: up to 56x faster TTFT on 130k-token prompts. Long-context VRAM limits ease.
Source ↗ - Society
🤖 15 European countries, 100+ groups in 2-day protest urge pause on large AI data centers
100+ groups in 15 European states plan 2-day demos Fri-Sat to pause AI/data-center projects. For infra investors, social/environmental costs explicit, raising siting/approval risk.
Source ↗
+8 more headlines
- 🤖 Instagram launches Edits for macOS with AI agent video editing
- 🤖 Baidu open-sources 280B ERNIE 5.5 Flash Base, beats DeepSeek on several benchmarks
- 🤖 OpenAI launches GPT-6 Cyber: outperforms GPT-6 Astra on most cybersecurity tasks
- 🤖 ElevenLabs integrates SoundCloud: one-click AI music release, 3 months Pro free
- ⚡ Aina launches conversational AI agent smart ring: with screen, joystick and mic
- 🤖 Zhang Yiming tops Forbes 2026 China Rich List for first time; Liang Wenfeng third, $46B
- 🤖 From 'cognitive fragments' to 'cognitive capital': AI boosts complex-task efficiency
- 🤖 AI automation decouples work from income: jobs, ownership, retraining as policy focus
Podcast highlight
It’s not sexy, nobody shares it, but it’s the kind of work that makes “cheap” actually real.