GLM-5.2

by Zhipu AI (Z.ai)

744B open-weight model, June 2026 — the release that made “China’s answer to Claude” a mainstream framing, and runnable on consumer hardware.

Part of the GLM-5 series.

Timeline

DateEvent
2026-06-05First appears in release-roundup coverage
2026-06-13Released; early-access testing widely published
2026-06-29Benchmarked head-to-head against Opus 4.8 and GPT-5.5 in GGUF form
2026-07-02”GLM-5.2 + GPT-5.5” promoted as a leading coding workflow
2026-08-01744B parameters running on consumer hardware, no GPU, via the Colibri engine
2026-08-18Adopted as a default free research model by practitioners

Notes

GLM-5.2’s significance is accessibility at scale. A 744B model is far beyond normal local reach,
yet the August coverage centres on running it on consumer hardware without a GPU — the Colibri
engine work is what made that plausible, and it reframed what “local” means for large models.

Pricing pressure is the recurring theme: “almost free” against Claude-tier capability. The
practical pattern that emerged was hybrid — GLM-5.2 for bulk generation, a frontier commercial
model for hard passes.