GLM-5

Overview

GLM-5 series includes versions 5, 5.1, and 5.2 with up to 744B parameters. It targets complex systems engineering and long-horizon agentic tasks, featuring solid 1M token context and advanced coding capabilities. The model achieves state-of-the-art performance on coding and agentic benchmarks.

README Preview

# GLM-5.2 & GLM-5.1 & GLM-5\n\n\n\n\n\n    👋 Join our Wechat or Discord community.\n    \n    📖 Check out the GLM-5.2 blog and GLM-5 Technical report.\n    \n    📍 Use GLM-5.2 API services on Z.ai API Platform. \n    \n    🔜 Try GLM-5.2 at z.ai.\n\n\n## Introduction\n\n### GLM-5.2\n\nGLM-5.2, our latest flagship model for long-horizon tasks. It marks a substantial leap in long-horizon task capability over its predecessor GLM-5.1 and, for the first time, delivers that capability on a **solid 1M-token context**. \n\nGLM-5.2's new capabilities include:\n- **Solid 1M Context:** A solid 1M-token context that stably sustains long-horizon work\n- **Advanced Coding with Flexible Effort**: Stronger coding capabilities with multiple thinking effort levels to balance performance and latency\n- **Improved Architecture**: We propose [IndexShare](https://arxiv.org/abs/2603.12201), which reuses the same indexer across every four sparse attention layers, reducing per-token FLOPs by 2.9× at a 1M context length. We also improve GLM-5.2’s MTP layer for speculative decoding, increasing the acceptance length by up to 20%\n\n\n\nOn standard coding benchmarks, GLM-5.2 is the strongest open-source model, improving on GLM-5.1 by a wide margin: 81.0 vs. 62.0 on Terminal-Bench 2.1 and 62.1 vs. 58.4 on SWE-bench Pro. It also closes much of the gap to the closed-source frontier — on Terminal-Bench 2.1 (81.0) it lands within a few points of Claude Opus 4.8 (85.0) — while staying ahead of Gemini 3.1 Pro.\n\nFor more detail, check our [blog](https://z.ai/blog/glm-5.2).\n\n### GLM-5.1\n\nGLM-5.1 is our next-generation flagship model for agentic engineering, with significantly stronger coding capabilities than its predecessor. It achieves state-of-the-art performance on SWE-Bench Pro and leads GLM-5 by a wide margin on NL2Repo (repo generation) and Terminal-Bench 2.0 (real-world terminal tasks).\n\n\n\nBut the most meaningful leap goes beyond first-pass performance. Previous models—including GLM

FAQ (3)

import re

def clean_reasoning(text: str) -> str:
    text = re.sub(r'(?i)(?:\bhmm?[\s,.!?-]*){2,}', '', text)
    text = re.sub(r'\s+', ' ', text)
    return text.strip()

reasoning = response["choices"][0]["message"].get("reasoning_content", "")
shown = clean_reasoning(reasoning)