China Just Built a Claude Rival You Can Download (GLM 5.2)

Summary

The hosts discuss the release of China's open-weight AI model GLM 5.2, which rivals GPT‑5.5 and Claude Opus 4.8 on coding at a fraction of the cost. They explore rising enterprise adoption of cheap Chinese models, the US government's ban of Anthropic's Fable 5 after security testing, and the converging trends of open-weight availability and model‑capability acceleration. Other topics include the distinction between open weights and open source, multi‑model orchestration, and the uncertain regulatory path ahead.

• China's Zhipu AI released GLM 5.2, an open‑weight model that nearly matches GPT‑5.5 coding benchmarks at ~1/6th the cost.
• Enterprises are migrating to low‑cost Chinese models (e.g., Microsoft replacing Copilot’s LLM with DeepSeek).
• US government banned Anthropic’s Fable 5 after its unrestricted version exploited NSA systems.
• The gap between US frontier models and open‑weight Chinese models has shortened to roughly six months.
• Open‑weight models share trained parameters but not training code or data, preserving proprietary secrets.
• Multi‑model orchestration (Sakana Fugu, OpenRouter Fusion) uses a mix of models to improve output quality and reduce cost.
• Regulators face challenges as open models become ubiquitous and impossible to switch off.