Correspondent · Global Labs
Mei-Lin follows the AI ecosystem beyond Silicon Valley — the Chinese labs, European challengers, and the global race for capability.
Anthropic says code merged into its codebase jumped 8x in 2026 compared to 2021–2024, which it views as preliminary evidence that AI is already beginning to accelerate the development of better AI.
Alibaba's Qwen team introduces Qwen3Guard, its first safety guardrail model for real-time AI moderation, alongside new image editing and translation tools.
Qwen3Guard is the first safety guardrail model in Alibaba's Qwen family, built on Qwen3 and designed to score risks on prompts and responses in real time across English, Chinese, and multilingual settings.
New research shows AI systems reliably outperform even elite human persuaders in text-based conversations, including raising real money for charity, with implications for how influence will be distributed in society.
Veterans from the UK AI Security Institute and startup Timaeus have formed Sequent, a nonprofit pursuing high-confidence alignment techniques for superintelligence, with plans to raise $100-150M initially.
The open-weight GLM-5.2 model is the first to crack into top-tier coding agent performance alongside Anthropic and OpenAI's best — and it landed right after the U.S. effectively banned Claude Fable 5.
China and 29 nations formed the World Artificial Intelligence Cooperation Organization, a bid to build AI governance outside Western influence, announced alongside training slots for the Global South.
A Chinese startup has released the world's largest open AI model, directly competing with US frontier systems and sending AI stocks tumbling.
Moonshot unveiled an open AI model that rivals Anthropic and OpenAI's offerings, sending AI and semiconductor stocks sliding and signaling China's aggressive open-source bet.
Claude Sonnet 5, Grok 4.5, GPT-5.6 Sol/Terra/Luna, and Kimi K3 all shipped within three weeks. Here's the full timeline — what launched, what it costs, how it benchmarks, and what the wave means.
Moonshot AI's Kimi K3 is the largest open-weight model released yet — 2.8T parameters, a 1M-token context, and $3/$15 pricing. Here's how it benchmarks against Opus 4.8, GPT-5.6, and Grok 4.5, plus how to use it and when open weights land.
The desk