researchaudio

Zhipu's GLM-5 tops open-source benchmarks with a novel async RL framework called SLIME.

What changed in the API, what failed during agent testing, and how to design tool-using systems that stay inside the rails.

Anthropic's randomized trial with 52 developers. But some interaction patterns beat hand-coding.

GPT-5.3-Codex debugged its own training. It also triggered OpenAI's first "High" cybersecurity rating.

Anthropic's new flagship outperforms GPT-5.2 by 144 Elo on knowledge work. Here's the technical breakdown.