Daily Brief Archive

Samsung deploys ChatGPT and Codex to global employees, one of OpenAI's largest enterprise rollouts; Stanford white paper reveals 53.5% of DeepSeek's core researchers are purely domestically trained, challenging the 「returnee narrative」; embodied intelligence 'brain' sector raises 43.8 billion yuan in six months, with world models becoming the primary focus for nearly 80% of funded companies

John Jumper, Father of AlphaFold, Leaves Google to Join Anthropic, as Google Loses Two Top Talents in 48 Hours; GLM-5.2 Defeats Fable 5 in Design Arena, Becomes First MIT-Licensed Open Source Champion Model, Priced at Just 1/7 to 1/11 of Competitors; NVIDIA, CMU, and Berkeley Launch ENPIRE Framework, Enabling Multiple Coding Agents to Autonomously Complete Full Robotics Research Pipeline, Achieving 99% Success Rate in Pin Insertion Task Within 3 Hours

GLM-5.2 becomes the first open-source model to pass the frontier experience, outperforms Fable 5 on BridgeBench reasoning at 1/10 the cost; OpenAI recruits Transformer paper author Noam Shazeer to strengthen pre-IPO R&D; Midjourney ventures into medical hardware with a 60-second full-body ultrasound scanner, plans to deploy 50,000 units

OpenAI collaborates with hundreds of doctors across 60 countries to enhance ChatGPT's health capabilities, opens GPT-5.5 Instant for free, and o3 achieves 18 new diagnoses in 376 difficult cases; OpenAI's alignment research improves 44 of 53 independent evaluations; Perplexity launches self-evolving memory system Brain, increasing accuracy by 25% on familiar tasks and reducing costs by 13%

OpenAI teams with 173 scientists to launch LifeSciBench, GPT-5.4 completes full drug discovery pipeline with 88% yield improvement in boronic acid reactions; DeepSeek secures over $7.4 billion in funding, valuation surpasses $50 billion; Radical AI's autonomous lab synthesizes 1,200 alloys in 6 months, accelerating materials discovery 10x

SpaceX acquires Anysphere, parent company of Cursor, in $60B all-stock deal to enter AI programming market; OpenAI's 「Deployment Simulation」 method achieves 92% accuracy in predicting model safety behavior pre-release; NVIDIA Blackwell sweeps all benchmarks in MLPerf Training 6.0 and sets new training time records

Alibaba open-sources 8 models in Qwen3 series, flagship 235B surpasses DeepSeek R1 and approaches Gemini 2.5-Pro; SpaceX completes $857 billion record IPO, market cap tops $2.5 trillion next day; U.S. government bans foreign access on national security grounds, forcing Anthropic to globally remove Fable 5/Mythos 5

Zhipu's GLM 5.2 open-sources to fill regulatory gap left by Anthropic's Fable 5; Microsoft launches MAI model family cutting costs by 90% and reducing reliance on OpenAI; Meta's $2B acquisition of Manus AI halted by Beijing, highlighting AI technological nationalism

The U.S. government imposes its first direct restriction on AI model exports, forcing Anthropic to globally take down Fable 5 and Mythos 5; SpaceX completes the largest IPO in history with a $2.1 trillion valuation, making Musk the first trillionaire; Zhipu’s GLM-5.2 and Moonshot’s Kimi K2.7-Code go open-source consecutively, enabling domestic AI players to capture market gaps left by Claude’s removal