GLM 5.2, a 753B open-weight model from Zhipu AI, is benchmarking against Opus 4.8 on coding and agentic tasks — and anyone can download and modify it. DeepSeek's speculative decoding delivers 50–400% faster inference with no retraining. Plus: Ornet, a model that writes its own test scaffolding.

Source episodes:

- The $10B Satellite Empire Putting AI in Orbit, Why Chips Beat Rockets & China's #1 Open Model | EP #266 (Moonshots) https://traffic.megaphone.fm/DVVTS6893487949.mp3

- Ornith 1.0: This is new class of self-improving model (Engineer Prompt) https://www.youtube.com/watch?v=25j4kMGhKGw

- The Capability Overhang Playbook (The AI Daily Brief) https://anchor.fm/s/f7cac464/podcast/play/122094487/https%3A%2F%2Fd3ctxlq1ktw2nl.cloudfront.net%2Fstaging%2F2026-5-28%2F426966199-44100-2-33e95372fce36.mp3

- DeepSeek Just Made Every LLM Faster, For Free (Engineer Prompt) https://www.youtube.com/watch?v=eFgknPFK-g0