This post rounds up several recent projects that didn’t justify their own full Interconnects write-up. It explains why I invested time in them and what they achieved.
**The ATOM Report: Measuring the Open Language Model Ecosystem.** The RLHF Book is now complete and available for pre-order! I’m creating a post-training course. Recent technical paper. Sharing here: https://arxiv.org/abs/2604.07190 Alongside the ATOM Project memo—an informal manifesto advocating for greater U.S. investment in open models, first launched in August 2025—we have published an updated technical report featuring our latest data, analysis, and narrative on the open language model ecosystem. The ATOM Report is packed with the methods that Florian and I rely on to monitor the open ecosystem. It explores GPT-OSS’s rapid rise, shifts in inference market share, the growing impact of China’s mid-tier players such as Moonshot, Z.ai, and MiniMax, encouraging signs of U.S. progress on open models, and more. The paper particularly focuses on our refinements to the Relative Adoption Metric (RAM), a measure we use to assess recent models’ adoption in a time-varying, size-normalized way. Here’s a selection of recent models—mostly from China—along with their RAM scores.
Interconnects AI