GLM-5: China's First Public AI Frontier Model

Z.ai (Zhipu), China's first publicly traded foundation model company, officially released GLM-5 on February 11th, 2026. This new frontier large language model is a 744B-parameter Mixture-of-Experts (MoE) model with 40B active parameters per token, a significant scale-up from GLM-4.5. It utilizes DeepSeek Sparse Attention (DSA) for a 200K-token context window and was trained on 28.5T tokens. GLM-5 is released under the MIT license and is available via the Z.ai API and OpenRouter.

A key highlight is its training entirely on Huawei Ascend chips using the MindSpore framework, demonstrating zero dependency on NVIDIA hardware. This is a geopolitical statement, as Zhipu has been on the U.S. Entity List since January 2025, banning access to H100/H200 GPUs. GLM-5 achieved strong benchmark results, including 77.8% on SWE-bench Verified (beating Gemini 3 Pro and GPT-5.2) and a score of -1 on the AA-Omniscience Index, indicating improved hallucination resistance. Its internal CC-Bench-V2 suite showed a 26% improvement over GLM-4.7 on frontend tasks, signaling a shift towards "agentic engineering."

Before its official launch, GLM-5 was stealth-tested as "Pony Alpha" on OpenRouter, processing over 40 billion tokens on its first day. The official API is priced at $1.00 per million input tokens and $3.20 per million output tokens, significantly cheaper than Claude Opus 4.6. Zhipu's Hong Kong IPO on January 8th raised $558 million at a $7.1 billion valuation. However, GLM-5 is text-only, lacking native multimodal support found in models like Kimi K2.5 from Moonshot AI, and some benchmark methodologies have raised questions. It requires substantial infrastructure (e.g., 8 H200s for FP8 inference), making it primarily an API model for most users.