Frontier Collapse: How Chinese Open-Weight LLMs and Automated Exploits Reshape Executive Risk
The AI Frontier Frontier artificial intelligence development this week shifted decisively toward specialized efficiency and high-capacity agentic architectures. Google expanded its efficiency tier with Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, halving task completion times while preserving performance baselines. Simultaneously, frontier intelligence benchmarks faced disruption from open-weight releases, most notably Moonshot AI's Kimi K3, a 2.8-trillion-parameter model with a 1-million-token context window that achieved top-five rankings across coding and agent leaderboards. Infrastructure demand continues to outpace localized capacity, driving Meta into negotiations to lease $10 billion in data center computing power to Anthropic. At the software layer, agentic deployment accelerated rapidly as OpenAI logged 10 million weekly active users across Codex and ChatGPT Work following the rollout of GPT-5.6. These developments demonstrate that model scale is no longer restricted to c...