Kimi K3 dropped last week, and Nathan Lambert and Florian Brand used it as a forcing function to map the entire open-model landscape on Interconnects. The episode covers the US-China geopolitical split, the economics of open versus closed models, and the distillation debate that Lambert calls consistently misread. Xi's WAIC speech explicitly named open source as a national strategy. Qwen announced its next major model will be open-weight. The pace is not slowing.
The most useful part of this episode is not the verdict on Kimi K3. It is the 30-minute stretch from timestamp 17:41 to 44:12 where Lambert and Brand tour the Chinese lab ecosystem, Qwen, DeepSeek, MiniMax and others, then pivot to the US open-model ecosystem and make a specific cybersecurity argument against frontier model bans. They also dissect the Ben Thompson distillation debate directly, which has been polluting takes across the industry for weeks. Lambert's provocation: benchmark selection is being weaponized by both sides to answer the question of how many months open models lag behind closed ones, and almost everyone is pulling the wrong number.
The model weights for Kimi K3 were not public at recording time, with a stated release target of July 27. Lambert's bet is that fine-tuning K3 on narrow, high-value tasks will yield large gains fast, citing GLM 5.2 as the prior template. The catch: loading K3 requires a full node of B300s, which makes post-training it an engineering problem most teams cannot solve yet. If you want to understand why the open-closed gap debate is the wrong frame entirely, read the original.
[READ ORIGINAL →]