Daily Brief Archive

Marvell plans a $115 billion all-stock acquisition of Lumentum; Meituan releases LongCat-2.5 with 1.6 trillion total parameters; Deep Principle releases open weights for its 753-billion-parameter scientific model Hydrogen-1.0

Kling AI completed a RMB 19.8 billion capital increase, with Kingdee, MetaX, and other institutions participating; Colossus 2 plans to add around 420,000 GB300s, targeting 1.2 million Blackwell GPUs by the end of 2026; Anthropic launched Claude Opus 5.5, with a weekly industry report saying costs are down 40%.

DeepSeek’s second funding round is reportedly RMB 105.5 billion, valuing the company at approximately RMB 615.5 billion; DeepSeek’s Super-agent had an internal-test task time of 15.7 minutes at p95, compared with 29.5 minutes for the traditional single-Session approach; S&P lowered Meta’s credit outlook to negative, and Meta’s third-quarter free cash flow was negative US$2.12 billion

苹果与阿里将Qwen接入中国市场Apple Intelligence，11月4日开始测试、计划12月随iOS 27.3上线；OpenAI最新融资目标据报从300亿美元下调至200亿美元，估值预期降至1.25万亿美元；BBC报道称，Alphabet因AI搜索降低付费广告可见度而股价下跌约12%、市值减少约5000亿美元

Mistral开放万亿参数原生多模态Large 4权重并支持自行托管; OpenAI披露四层前沿模型监控体系，可对高风险任务自动暂停; X-Era开源E0.5，双臂与人形机器人平均成功率分别达85.0%和88.3%

Google发布Gemini 4 Carbon，漏洞发现、利用和修复得分分别为88.2%、54.8%和71.0%; GitHub报告显示月合并Pull Request由约7000万增至2.2亿，提交数升至73.8亿; SK海力士称未来三年DRAM与HBM产能基本售罄，并将显著提高2027年资本开支

Google发布Genie 4，可生成含音频的交互世界并维持最长两分钟一致性; OpenAI宣称AI解决三项基础科学问题，相关证明正等待外部学界验证; Liquid AI发布六款低于20亿参数的LFM3端侧多模态模型，可用单张GPU微调

OpenAI发布GPT-6.1 Terra，API定价降至每百万输入Token 1美元、输出Token 5美元; Anthropic提交IPO申请，上半年收入163亿美元、净亏损807亿美元; 腾讯开源Hy-SimulMT，中英同传响应约3秒

辉羲智能2026年内累计融资25亿元并转向具身智能芯片；韩国Hugging Face新模型中49.3%基于中国模型，Qwen占46.8%；Google开源ML Drift，Gemma基准内存开销降低12%