The Sequence Radar #897: Last Week in AI: China, Compression and the Open-Model Race

This week’s AI advancements shifted focus from raw scale to questions of access, governance, and distribution. New models like Thinking Machines’ Inkling and Moonshot AI’s Kimi K3 emphasize open weights and user control, while PrismML’s Bonsai 27B demonstrates radical compression for edge deployment. OpenAI’s GPT-Red showcases a new scaling law for safety through automated red-teaming, hinting at machine-speed evolutionary ecosystems.

• Thinking Machines released Inkling, a 975-billion-parameter open-weight model with a 1M context window, focusing on customization.
• Moonshot AI introduced Kimi K3, a 2.8 trillion-parameter model with open weights, aiming for long-horizon coding.
• PrismML’s Bonsai 27B offers extreme compression, with a 5.9GB model small enough for smartphones.
• OpenAI’s GPT-Red is an automated red-teaming system that uses self-play to find vulnerabilities, significantly improving model robustness.
• Xi Jinping framed open-source AI as a global public good, highlighting China’s strategic push for AI governance and international cooperation.
• The trend indicates a fragmentation and potential democratization of AI, shifting the contest from building the smartest system to deciding whether intelligence will be centralized or portable.
• Apple Intelligence is approved for launch in China, integrating Alibaba’s Qwen models.
• Emergent announced a $130 million Series C funding round.
• Demis Hassabis proposed an independent standards body for reviewing frontier models.
• Reflection AI secured a $1 billion+ compute deal with Nebius.
• Google’s Gemini 3.5 Pro is reportedly behind schedule.
• Greylock launched an $1.5 billion early-stage fund.
• TSMC’s revenue shows continued AI demand.
• Walden Robotics launched with significant seed funding.
• SK Hynix completed a record-breaking US listing.
  https://bender.layer3.press/articles/9216a6fc-4a6e-4613-9ded-d707730b3727