Daily Brief Archive

Anthropic's annualized revenue surges from $9 billion to $30 billion, surpassing OpenAI, and secures 3.5GW of Google TPU compute; Zhipu's GLM-5.1, a 754B-parameter MIT-licensed open-source model, outperforms GPT-5.4 and Opus 4.6 on SWE-bench Pro; Anthropic refrains from public release of Claude Mythos due to its potent cyberattack capabilities, launching Project Glasswing with 40+ partners for defensive collaboration

DeepSeek V4 opensource trillion-parameter MoE model with only $5.2M training cost; Nvidia Rubin GPU production may drop 25% due to HBM shortage, Microsoft and Google sign three-year DRAM contracts to secure supply; OpenAI faces internal IPO dispute as CFO questions feasibility amid $600 billion spending commitment

GPT-6 codenamed Spud completes pretraining, context window potentially reaching 2 million tokens with 40% gains in coding and reasoning; China's Cyberspace Administration releases draft regulations for digital virtual humans, banning provision of virtual intimate relationship services to minors; NVIDIA invests billions in photonic interconnect technology, planning to build 1,000-GPU ultra-scale systems by 2028

Anthropic blocks third-party subscription access to OpenClaw, developers must switch to paid API; Arcee AI releases 399B open-source reasoning model, output cost just 4% of Claude Opus; NVIDIA executive warns GPU supply will be very tight in coming months, half of U.S. AI data center projects face delays

Anthropic acquires Coefficient Bio for $400M to enter life sciences AI; US VC funding reaches record $267.2B in Q1 2026 with AI占比89%; AI-generated security reports undergo qualitative shift, Linux kernel receives 5–10 high-quality vulnerability reports daily

Google releases four Gemma 4 open source models under Apache 2.0 license, 31B ranks third globally; Microsoft launches three proprietary MAI models with speech-to-text word error rate at 3.8%, lower than OpenAI and Google; foundational AI funding in Q1 2026 reaches $178 billion, double the 2025 annual total

Liquid AI releases 350M-parameter model LFM2.5-350M with 80,000:1 token-to-parameter ratio enabling efficient edge inference; NVIDIA Blackwell Ultra sets records in MLPerf v6.0, achieving 2.7x performance gain on DeepSeek-R1; SpaceX secretly files IPO application targeting $1.75 trillion valuation

OpenAI completes $122 billion in historic financing, valued at $852 billion, and shuts down Sora; Claude Code leaks 500,000 lines of source code due to npm misconfiguration, exposing unreleased features like KAIROS memory system; Microsoft Copilot integrates GPT+Claude multi-model Critique, boosting accuracy by 13.8%

Google releases Gemini 3.1 Pro, scores 77.1% on ARC-AGI-2 to reclaim title of strongest model; Reflection AI valuation jumps 46x to $25 billion while seeking $2.5 billion funding; Insilico Medicine and Lilly announce $2.75 billion AI drug discovery collaboration triggering antitrust review