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

Summary

A podcast about Kimi K3 and China's open models examines the performance gap, distillation debates, and geopolitical impacts.

Why it matters

The lag of open models behind closed models by months is a critical issue for global AI competition and security policies. Kimi K3's ability to forecast trends through Reddit analysis highlights the innovation potential within the open model ecosystem.

Highlights

• Kimi K3 achieves performance close to closed models on certain tasks, but requires high engineering and hardware resources.
• China's release of models like Qwen 3.8 with open weights is a strategic move within the US-China geopolitical context.
• The debate that the performance gap can vary by months depending on benchmarking metrics, and that distillation offers limited benefits.

Critical take

The inconsistency of benchmarks measuring the performance gap and Kimi K3's current difficulty in large-scale fine-tuning suggest it may remain limited in real-world applications.