GLM-5 is Z.ai's flagship open-source large language model, purpose-built for complex systems engineering and long-horizon agentic tasks. It is released openly so developers can self-host a frontier-class model with full deployment control. ## Why GLM-5 Matters Most open-weight models still trail closed frontier systems on the hardest agentic and coding work. GLM-5 closes much of that gap. On the team's internal CC-Bench-V2 suite it significantly outperforms its predecessor across frontend, backend, and long-horizon tasks, and it ranks as a best-in-class open-source model on reasoning, coding, and agentic benchmarks while narrowing the distance to closed frontier models. ## Architecture and Scale GLM-5 scales from GLM-4.5's 355B parameters (32B active) to 744B parameters (40B active), and increases pre-training data from 23T to 28.5T tokens. It integrates DeepSeek Sparse Attention (DSA), which substantially lowers deployment cost while preserving long-context capacity. Post-training relies on slime, an asynchronous reinforcement-learning infrastructure that improves training throughput and enables more fine-grained iteration. ## Built for Long-Horizon Agentic Work GLM-5 is designed to stay effective over long agentic sessions rather than plateauing after early gains. It breaks complex problems down, runs experiments, reads results, identifies blockers, and revises its strategy across many rounds. On Vending Bench 2, which measures long-term operational capability, GLM-5 ranks #1 among open-source models. ## The GLM-5 Family The repository hosts the full GLM-5 line. GLM-5.1 pushes agentic engineering further with state-of-the-art results on SWE-Bench Pro and large gains on repository generation and Terminal-Bench 2.0. GLM-5.2, the latest flagship, adds a solid 1M-token context, flexible coding effort levels, and the IndexShare architecture that reuses one indexer across every four sparse-attention layers to cut per-token FLOPs by 2.9x at 1M context. On Terminal-Bench 2.1 it scores 81.0, landing within a few points of the closed frontier while leading other open-source models. ## Access and Licensing GLM-5 is available under the Apache-2.0 license with open weights, alongside hosted API services on the Z.ai platform for teams that prefer managed inference.