--- title: "GLM-5: A Leading Open-Source Model for Long-Horizon Coding and Agentic Tasks" url: https://daily.dev/posts/glm-5-a-leading-open-source-model-for-long-horizon-coding-and-agentic-tasks-bospgaonw source_url: https://daily.dev/posts/glm-5-a-leading-open-source-model-for-long-horizon-coding-and-agentic-tasks-bospgaonw type: collection source: "Collections" published: 2026-02-11T18:08:07.749Z updated: 2026-02-15T14:26:14.702Z tags: ["ai", "machine-learning", "open-source", "llm"] reading_time: 2 upvotes: 1 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # GLM-5: A Leading Open-Source Model for Long-Horizon Coding and Agentic Tasks **[Collections](https://daily.dev/sources/collections)** · 2 min read · 1 upvotes · 0 comments ## Summary GLM-5 is a new open-source language model with 744 billion parameters (40 billion active) trained on 28.5 trillion tokens, released under MIT license. It uses DeepSeek Sparse Attention and asynchronous reinforcement learning to compete with proprietary models like Claude Opus and GPT in coding tasks. The model demonstrated autonomous coding by building a Game Boy Advanced emulator with GUI in 24 hours. Modal offers free API access through April, with deployment across multiple GPUs achieving 30-75 tokens/second, and integration with tools like OpenCode, Claude Code, and Vercel AI SDK. ## Content GLM-5, the latest open-source language model developed by Z.ai, marks a new milestone in the realm of AI technology. With its state-of-the-art features and robust performance, this model is designed to rival proprietary offerings such as Claude Opus 4.6 and GPT 5.3 Codex, particularly in long-horizon coding tasks. Released under the MIT license, GLM-5 boasts 744 billion parameters with 40 billion active ones, trained extensively on 28.5 trillion tokens. One of its remarkable attributes is the implementation of DeepSeek Sparse Attention, which is enhanced by a unique asynchronous reinforcement learning infrastructure known as ‘slime.’ These innovations enable the model to excel in reasoning, coding, and agentic tasks, achieving the top rank in Vending Bench 2 for long-term operational capacity. The model has captured attention through its impressive autonomous coding capabilities. For instance, it independently created a functional Game Boy Advanced emulator with a graphical user interface within a mere 24 hours, relying solely on a system prompt and hardware documentation. This feat underscores its potential to perform complex, diverse tasks autonomously and highlights the rapidly evolving potential of open-source AI models. Currently, Modal provides free access to GLM-5 via an OpenAI-compatible API endpoint until the end of April. Users can deploy this massive 700GB model across multiple GPUs using SGLang, achieving a processing speed of 30–75 tokens per second per user. Additionally, it integrates seamlessly with various popular AI tools such as OpenCode, OpenClaw, Claude Code, Vercel AI SDK, and even coding agents like Cline, supporting document generation and cross-application automation. In summary, GLM-5 represents a significant leap forward in open-source AI technology, showcasing capabilities that challenge the top proprietary models. Its impactful contributions to AI-driven coding and automation offer a glimpse into the future possibilities of intelligent, autonomous systems. --- Tags: [#ai](https://daily.dev/tags/ai), [#machine-learning](https://daily.dev/tags/machine-learning), [#open-source](https://daily.dev/tags/open-source), [#llm](https://daily.dev/tags/llm) [View this post on daily.dev](https://daily.dev/posts/glm-5-a-leading-open-source-model-for-long-horizon-coding-and-agentic-tasks-bospgaonw)