Qwen3.7-Max Runs Solo 35h, Cohere Drops 218B MoE—AI Daily Brief
Audio in Mandarin Chinese · English transcript below
⚡ Alibaba's 35-hr AI marathon, Cohere cranks 220B on 2 GPUs, Waymo rain-out in 4 cities, OpenAI kills Mars coffee, India phones pack spicy chat, US-China code duel rages!
Today's Top 3 Headlines
- AI Industry News
🤖 Qwen3.7-Max runs 35h, 10× speed, $2.08M virtual revenue
VentureBeat: Qwen3.7-Max ran 35 hrs, 1,158 calls on foreign chips, 10× code speed-up, $2.08 M sim-cash. Devs gain tireless long-horizon Agent; investors eye fresh compute ROI.
Source ↗ - AI Product Launch
🚀 DeepSeek Code Harness; Beijing wants full-stack AI
Insiders: DeepSeek coding IDE Code Harness fills its AI stack. One more domestic dev chain link; a key Beijing compute-data loop move.
Source ↗ - AI
🤖 Ex-Tencent AI chief Liu Wei: China LLMs trail, but game still winnable
China’s LLMs lag OpenAI by 1-2 gens, Liu Wei tells 36Kr; vertical apps & data edge can still win. Founders should fund scenario-specific, hardware-software combos, not chase parameters.
Source ↗
+9 more headlines
- 🤖 Starbucks Halts NA AI Inventory Roll-Out Amid Ops Hurdles
- 🤖 Cohere drops 218B sparse MoE Command A+, runs on 2 H100s
- 🤖 Waymo halts 4-city ops; robo-taxis still wade floods
- ⚖️ Chris Lehane tackles OpenAI's global PR crisis
- 🤖 HMD Vibe 2 5G ships with 22-lang Indus AI, ₹10,999
- 🤖 OpenMythos Deep Dive: Tuning Looped-Depth Transformer + MLA/GQA Playbook
- 🤖 CODA rewrites Transformer blocks as GEMM-Epilogue, letting LLMs code kernels
- 🤖 SpaceX, OpenAI, Anthropic race to IPO; AI capital map to be redrawn
- 🤖 DeepSeek builds Harness team to rival Claude Code in coding Agent race