What this is
Zhipu AI (rebranded Z.ai internationally) shipped the most consistent quarterly release cadence among Chinese labs in 2024-2026. GLM-5 (February 2026) is a 744B-parameter MoE that made Z.ai the first publicly-listed Chinese AI lab (HKEX, January 2026). This page is a 2026-05-15 release-by-release reference.
GLM Release Timeline (2024-2026)
| Date | Release | Notable |
|---|---|---|
| Jun 5, 2024 | GLM-4 (incl. GLM-4-9B open) | Foundation release |
| Apr 2025 | GLM-4-32B-0414 series | Scaled to 32B; dialogue/reasoning/rumination variants |
| Jul 2025 | GLM-4.5 + GLM-4.5 Air | Next-gen language; runs on 8x NVIDIA H20 |
| Aug 11, 2025 | GLM-4.5V | 106B vision-language model |
| Sep 2025 | GLM-4.6 | First integration of FP8 + Int4 on Cambricon chips |
| Dec 2025 | GLM-4.6V + GLM-4.7 | Vision + agentic coding tier |
| Jan 8, 2026 | Z.ai HKEX IPO | First publicly listed Chinese AI lab |
| Feb 11, 2026 | GLM-5 | 744B MoE / 40B active; 2x scale-up from 4.5 |
| Mar 2026 | GLM-5.1 (subscription) | Refinement release |
| Apr 8, 2026 | GLM-5.1 open-source | Open-weight release |
GLM-5 Specifications
| Spec | Value |
|---|---|
| Total parameters | 744B |
| Active parameters per token | 40B |
| Architecture | Sparse MoE |
| Scale change vs GLM-4.5 | ~2x total params, ~1.25x active |
| License | GLM-5.1 open-sourced April 8, 2026 |
Six Things the Lineage Tells You
- Z.ai is the first publicly listed Chinese AI lab. HKEX IPO January 8, 2026 changes the capital structure and disclosure cadence of the company.
- GLM ships ~quarterly. 8+ major releases in 18 months — the most consistent cadence among Chinese frontier labs.
- GLM-4.6 was the first Chinese flagship to run on Cambricon chips at FP8 + Int4. Validated the Chinese chip stack at production scale.
- GLM-5 is a 2x scale-up. 744B total / 40B active vs 355B / 32B for GLM-4.5. Closing the gap to the largest open-weight frontier (Llama 4, DeepSeek V4).
- Z.ai rebrand internationalised the company. Zhipu retained domestically; Z.ai is the international face.
- GLM-5.1 open-sourcing (Apr 8, 2026) keeps Z.ai in the open-weight game. Distinct from labs that have moved entirely to closed-tier flagship.
What This Means for AI Visibility
Z.ai's public-company status, hardware-stack diversity (NVIDIA + Cambricon), and consistent open-source cadence make GLM the most institutionally-deployable Chinese open-weight model. Brand-visibility teams targeting enterprise deployments inside China should treat GLM as the most likely default open-weight model, alongside Qwen and DeepSeek.
Methodology
Release dates from Wikipedia Z.ai article, HuggingFace blog on GLM-5, Z.ai GLM-4 GitHub, Z.ai homepage, and RecodeChinaAI on the IPO.
How Presenc AI Helps
Presenc AI tracks brand visibility on GLM-served surfaces (Z.ai cloud API, chat.z.ai, on-prem GLM-5.1 deployments). As GLM-5 + GLM-5.1 absorb enterprise share, brand teams need GLM-specific monitoring distinct from Qwen and DeepSeek baselines.