GLM

GLM is the parent name for a family of open-weights language models from the Chinese lab Z.ai (ex-Zhipu AI), with the dated releases GLM-5.1, GLM-5.2 and GLM-5.3 sitting under it as variants.

What the source notes return to is less the architecture than the closing gap. Artificial Analysis, on 22 June 2026, placed GLM-5.2 at 1524 Elo on GDPval-AA, third overall and well clear of the next open model, MiniMax-M3, at 1408. Only Claude Fable 5 (1783) and Claude Opus 4.8 (1615) scored higher, and GLM-5.2 tied GPT-5.5 at xhigh (1509). The model is a 753B-parameter Mixture of Experts with 40B active, MIT licensed, priced at $1.40 / $4.40 per million tokens. Andrew Ng's The Batch n°350 reported GLM-5.1 at the top of SWE-Bench Pro (58.4%) while trailing on reasoning, GPQA Diamond 86.2% against Gemini 3.1 at 94.3%.

The August 2026 GLM-5.3 announcement moved the story toward security. Z.ai reported that, working with Chinese security teams, the model found 2,436 vulnerabilities across 269 open source projects, average lifetime 26.6 years, with 53 disclosed and 2,383 under embargo. Weights were promised "in two weeks, once safety evaluation and hardening are complete": a date, not a criterion.

Eric Schmidt, on 21 October 2026, framed the stakes differently: China is roughly six months behind, and its free models will win users in poor countries. He left the question of removing guardrails from open models unresolved.