三星被曝收缩中国手机布局：月销不足30万元的门店，将逐步清退；宗馥莉的「同父异母兄弟」成立新公司；约120万辆特斯拉汽车遭调查
Ranking
Overall
29
Content
20
Popularity
N/A
No observed public metrics; popularity remains neutral/archived.
Merged summary
TL;DR - A 雷峰网 daily briefing whose AI-relevant items center on China's model race: Alibaba's Qwen3.8-Max launch, DeepSeek's surging token throughput forcing OpenAI price cuts, and MiniMax's H3 omni-modal open-source release. Together they show pricing and open weights, not just capability, driving the current competitive front.
- Alibaba released Qwen3.8, a 2.4T-parameter base model targeting coding and office work, with API pricing at ¥12/M input and ¥36/M output (¥1.5 on implicit cache hits) — claimed at 40%/24% of Opus 5's international price; Qwen3.8-Max and Qwen3.8-27B are slated to open-source next week, and HK shares rose 7.01%.
- Alibaba also opened public beta of "千问办公" (QwenWork), pitched as the first product spanning desktop, cloud, and enterprise-collaboration agents, already partially wired into DingTalk and built by consolidating QoderWork, 悟空, and MuleRun.
- DeepSeek V4 Flash reportedly processed 8T tokens in one day on OpenCode (5T free-tier, 3T paid) and topped OpenRouter's weekly rankings at 7.22T tokens (+13% WoW), with Chinese models taking the top five slots; OpenAI responded by cutting GPT-5.6 Luna output pricing 80% ($6→$1.2/M) and input $1→$0.2/M roughly three weeks after launch.
- MiniMax open-sourced H3, an omni-modal system (H3-Context-IR, H3-Base, H3-Regenerate-2K) generating up to 2K/15s video with native stereo audio, with Day-0 support from 16 chip vendors and platforms including Ascend, Moore Threads, AMD, Intel, Hugging Face, and ComfyUI.
Sources (1)
三星被曝收缩中国手机布局：月销不足30万元的门店，将逐步清退；宗馥莉的「同父异母兄弟」成立新公司；约120万辆特斯拉汽车遭调查
Public signals
N/A
TL;DR - A 雷峰网 daily briefing whose AI-relevant items center on China's model race: Alibaba's Qwen3.8-Max launch, DeepSeek's surging token throughput forcing OpenAI price cuts, and MiniMax's H3 omni-modal open-source release. Together they show pricing and open weights, not just capability, driving the current competitive front.
- Alibaba released Qwen3.8, a 2.4T-parameter base model targeting coding and office work, with API pricing at ¥12/M input and ¥36/M output (¥1.5 on implicit cache hits) — claimed at 40%/24% of Opus 5's international price; Qwen3.8-Max and Qwen3.8-27B are slated to open-source next week, and HK shares rose 7.01%.
- Alibaba also opened public beta of "千问办公" (QwenWork), pitched as the first product spanning desktop, cloud, and enterprise-collaboration agents, already partially wired into DingTalk and built by consolidating QoderWork, 悟空, and MuleRun.
- DeepSeek V4 Flash reportedly processed 8T tokens in one day on OpenCode (5T free-tier, 3T paid) and topped OpenRouter's weekly rankings at 7.22T tokens (+13% WoW), with Chinese models taking the top five slots; OpenAI responded by cutting GPT-5.6 Luna output pricing 80% ($6→$1.2/M) and input $1→$0.2/M roughly three weeks after launch.
- MiniMax open-sourced H3, an omni-modal system (H3-Context-IR, H3-Base, H3-Regenerate-2K) generating up to 2K/15s video with native stereo audio, with Day-0 support from 16 chip vendors and platforms including Ascend, Moore Threads, AMD, Intel, Hugging Face, and ComfyUI.