{"id":51451,"date":"2025-08-07T23:42:14","date_gmt":"2025-08-07T23:42:14","guid":{"rendered":"https:\/\/www.newsbeep.com\/uk\/51451\/"},"modified":"2025-08-07T23:42:14","modified_gmt":"2025-08-07T23:42:14","slug":"glm-4-5-launches-with-strong-reasoning-coding-and-agentic-capabilities","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/uk\/51451\/","title":{"rendered":"GLM-4.5 Launches with Strong Reasoning, Coding, and Agentic Capabilities"},"content":{"rendered":"

Zhipu AI has released<\/a> GLM-4.5 and GLM-4.5-Air, two new AI models designed to handle reasoning, coding, and agent tasks within a single architecture. They use a dual-mode system to switch between complex problem-solving and faster responses, aiming to improve both accuracy and speed.<\/p>\n

GLM-4.5 features 355B total parameters with 32B active, while its lighter sibling, GLM-4.5-Air, runs with 106B total and 12B active parameters. Both models use a Mixture-of-Experts (MoE) architecture and are optimized for two modes: a \u201cthinking\u201d mode for complex reasoning and tool use, and a \u201cnon-thinking\u201d mode for fast responses.<\/p>\n

GLM-4.5\u2019s architecture prioritizes depth over width \u2014 a contrast to models like DeepSeek-V3 \u2014 and uses 96 attention heads per layer. It also incorporates QK-Norm, Grouped Query Attention, Multi-Token Prediction, and the Muon optimizer for faster convergence and improved reasoning performance.<\/p>\n

Training was conducted on a 22T-token corpus, including 7T tokens dedicated to code and reasoning, followed by reinforcement learning with Zhipu AI\u2019s in-house slime RL infrastructure. This setup features an asynchronous agentic RL training pipeline designed to maximize throughput and support long-horizon tasks.<\/p>\n

Zhipu AI reports that GLM-4.5 ranks 3rd overall on a combined set of 12 benchmarks covering agentic tasks, reasoning, and coding, trailing only the very top models from OpenAI and Anthropic. GLM-4.5-Air ranks 6th, outperforming many models of similar or larger scale.<\/p>\n



\nSource: Zhipu AI Blog<\/p>\n

GLM-4.5 demonstrated particular strength in coding benchmarks. It achieved 64.2% on SWE-bench Verified and 37.5% on TerminalBench, placing it ahead of Claude 4 Opus, GPT-4.1, and Gemini 2.5 Pro on several metrics. Its tool-calling success rate reached 90.6%, outperforming Claude-4-Sonnet (89.5%) and Kimi K2 (86.2%).<\/p>\n

Early testers have praised GLM-4.5\u2019s coding and agentic capabilities. One Reddit user shared<\/a>:<\/p>\n

<\/p>\n

These models seem extremely good from my preliminary comparison. GLM-4.5 seems excellent at coding tasks, while GLM-4.5-Air seems even better than Qwen 3 235B-a22b 2507 on my agentic research and summarization benchmarks.<\/p>\n

<\/p>\n

Another user commented<\/a> on the GLM series\u2019 speed and language proficiency:<\/p>\n

<\/p>\n

GLM is pretty impressive. Didn\u2019t try 4.5 yet, but 4.1 Thinking Flash scored around 150\/200 on Scolarius in French language testing \u2014 one of the best in my personal 19 LLM comparison. Extremely fast too.<\/p>\n

<\/p>\n

GLM-4.5 can be accessed directly via Z.ai, called through the Z.ai API<\/a>, or integrated into existing coding agents like Claude Code or Roo Code. Model weights for local deployment are available on Hugging Face<\/a> and ModelScope, with support for vLLM and SGLang inference frameworks.<\/p>\n","protected":false},"excerpt":{"rendered":"Zhipu AI has released GLM-4.5 and GLM-4.5-Air, two new AI models designed to handle reasoning, coding, and agent…\n","protected":false},"author":2,"featured_media":51452,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[31],"tags":[23863,554,733,6225,6485,6486,28452,1120,96,28451,13290,13289,56,54,55],"class_list":["post-51451","post","type-post","status-publish","format-standard","has-post-thumbnail","category-arts-and-design","tag-agents","tag-ai","tag-artificial-intelligence","tag-arts","tag-arts-and-design","tag-artsanddesign","tag-benchmark","tag-design","tag-entertainment","tag-glm-4-5","tag-large-language-models","tag-ml-data-engineering","tag-uk","tag-united-kingdom","tag-unitedkingdom"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/51451","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/comments?post=51451"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/51451\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media\/51452"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media?parent=51451"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/categories?post=51451"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/tags?post=51451"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}