GLM 5.1

GLM-5.1 is a 744B-parameter Mixture-of-Experts model developed by Zhipu AI (rebranded as Z.AI), released on April 7, 2026. It represents the current pinnacle of open-weight coding and agentic AI, achieving the highest public score on SWE-Bench Pro (58.4%) — edging past both GPT-5.4 and Claude Opus 4.6. The model was trained entirely on 100,000 Huawei Ascend 910B chips with zero NVIDIA hardware, making it a landmark achievement in China's AI self-sufficiency under US export controls. Z.AI completed a Hong Kong IPO in January 2026 at a ~$31.3B valuation, becoming the world's first publicly traded foundation model company.

Architecture-wise, GLM-5.1 is a post-training refinement of GLM-5, sharing the same MoE backbone (744B total, 40B active per token, 256 routed experts with top-8 selection) but with significantly enhanced coding, agentic, and long-horizon capabilities. Its defining innovation is extended autonomous execution: the model can sustain a plan-execute-analyze-optimize loop for up to 8 hours without human checkpoints, demonstrated by building a complete Linux desktop environment through 655 iterations and optimizing a vector database to 6.9× its original throughput. The model ships under the MIT license with weights available on Hugging Face.

Where GLM-5.1 falls short is general-purpose reasoning and multimodal tasks. It trails frontier closed-source models by 5-12 points on GPQA Diamond (86.2%), Humanity's Last Exam (31.0 base), and AIME 2026 (95.3%). It is text-only, with no image or audio processing. Generation speed (~40-44 tok/s) is below the competitive median, and the model's verbose output style inflates token usage. For organizations building autonomous coding agents or long-running engineering pipelines, GLM-5.1 is the strongest open-weight option available; for multimodal workflows or maximum reasoning capability, closed-source alternatives remain superior.