Open models recap: more on Kimi K3, Qwen 3.8, Xi's WAIC speech, distillation, the open-closed gap, and what's next

Details

Following Kimi K3's release, Nathan Lambert and Florian Brand recorded a 49-minute breakdown of the current open-model landscape for the Interconnects podcast. Two headline developments set the tone: Xi Jinping used a WAIC speech to directly commit to openness and open source as a national AI strategy — described by Lambert as more of a directional statement than a detailed policy — and Alibaba's Qwen announced its next major model will be open-weight, a shift Lambert calls a 'big change.' The conversation covers GLM 5.2's continued relevance, a roundup of Chinese providers (Qwen, DeepSeek, MiniMax), and a tour of the US open-model ecosystem. Florian Brand pushes back on benchmark-based 'months behind closed frontier' claims, arguing that proliferating benchmark providers and heavy benchmark optimization make precise gap estimates unreliable. Lambert adds that a post-training knowledge gap between Western and Chinese practitioners is underappreciated. The distillation section addresses Ben Thompson's framing of distillation as a value-destroying force for frontier labs. Lambert disagrees with the oversimplification. The episode closes with frontier tier list predictions. Content here is podcast analysis — no primary benchmarks or proprietary data are shared.