⌘K
Change language Switch ThemeSign In
Narrow Mode
GLM-5.1 上线，编程表现贴 Opus 4.6 开大，Coding plan 瞬间断货
量 量子位 @梦瑶
One Sentence Summary
Zhipu AI releases the GLM-5.1 model, featuring significantly improved programming capabilities that rival Claude Opus 4.6, with support for reasoning mode and integration with various programming tools.
Summary
This article reports on the surprise launch of Zhipu AI's GLM-5.1 model. This version sees a nearly 10-point increase in programming capability compared to its predecessor, with performance approaching the world-leading Claude Opus 4.6. The article details the model's key parameters (approx. 200K context, support for reasoning mode, OpenAI-compatible API), showcases real-world test results from users in scenarios like game development and spatial understanding, and provides a step-by-step guide for configuring GLM-5.1 in tools like Claude Code. Additionally, the article analyzes Zhipu's rapid iteration pace, noting that this version focuses on strengthening stability, execution efficiency, and performance in complex tasks.
Main Points
* 1. GLM-5.1 achieves a leap in programming capability, with performance rivaling industry benchmarks.In official Coding Evaluation benchmarks, GLM-5.1 shows a nearly 10-point improvement over the previous generation, trailing the currently recognized top programming model, Claude Opus 4.6, by only 2.6 points. * 2. The model's parameters and functional positioning are clear, focusing on complex code and reasoning scenarios.It features a 200K context window, supports reasoning mode, and provides an OpenAI-compatible interface, significantly lowering the barrier to entry for developers. * 3. Real-world testing demonstrates the model's evolution in spatial consistency and dynamic completion.Through real-world test cases like an AI-powered 'Minecraft' and interior design blueprints, it showcases the model's advantages in handling continuous environments, spatial structure understanding, and long-text logic generation. * 4. Provides highly valuable tool integration solutions.It details the specific steps to integrate GLM-5.1 into Claude Code by modifying the settings.json configuration file, enabling developers to quickly incorporate it into their production workflows.
Metadata
AI Score
87
Website qbitai.com
Published At Today
Length 2312 words (about 10 min)
Sign in to use highlight and note-taking features for a better reading experience. Sign in now
> 梦瑶 发自 凹非寺 > > > 量子位 | 公众号 QbitAI
智谱GLM-5.1，突然上线！
官方公告只有寥寥两句：
> GLM-5.1现已面向GLM Coding Plan全部用户(Lite/Pro/Max)开放。 > > ModelKey：GLM-5.1
别的没再多说，只是默默甩出Coding Evaluation评测结果——
在编程能力上相比上一代GLM-5直接飙升近10分。
甚至嘛，距全球最强编程模型Claude Opus 4.6，也就只有2.6分之差？？
虽说榜单上的对比模型不多，但架不住网友脑补空间大啊，这下大家哪还能坐得住：
网友：既然GLM-5.1的评分都快贴到Opus脸上了，那岂不是意味着…
很可能已经超过——
目前仅次于Opus 4.6的Sonnet4.6模型了？？
要是这样的话…那网友们直接反手就是一个实测大整活儿！！
看下面这位小哥，连夜搓出来了个AI版「我的世界」，还能带交互的内种，有点大型游戏那味儿了啊：
还有网友直接把模型安排到自己日常工作流里了。
下面这位朋友直接把关于灵巧手的研究资料喂给了GLM-5.1后，就得到了一本专业的行业手册！
（良心评价：输出的效果忒意外，远超我的预期～）
此外值得一提的是，此次的GLM-5.1版本率先向GLM Coding Plan所有用户开放。
结果您猜怎么着，太爆了，直接爆到《售罄》了…（doge）
是的，模型也能卖断货了。
（正在狠狠排队候补ing！！！)
u1s1，除了官方给出的那份Coding Evaluation外，这次智谱貌似连文档都还没好好写……
但从现有资料文档和网友的实测效果中，我们也能对GLM-5.1的模型能力有个大概齐的了解～
先来说说大家最为关心的模型基本信息：
* 上下文窗口：大约在200K级别，基本延续了GLM-5的配置，在量级上没有太大变化。 * 模型能力：从知识库文档的标题中也不难得出——模型定位主要还是偏代码场景，此外支持reasoning模式。
* 模型配置：目前支持在Claude Code里手动切换使用，同时也能接入OpenClaw进行自定义配置。
* 接口兼容：支持OpenAI Compatible接入，接入入门槛基本被压到了最低，对开发者来说不用改太多代码就能让模型跑在工作流里～ * 模型开放对象：面向所有GLM coding plan用户。（是的，Lite用户也能用了～）
光说性能还是忒抽象！咱们直接拿已经抢先尝鲜网友们的实测效果，看看模型表现到底如何～
先来看看youtube上@AICodeKing网友大神的实测效果！
下面这个室内设计平面图就是小哥用GLM-5.1做出来的。
从输出效果来说，像厨房、客厅、卧室这些区域的划分都比较明确，甚至连动线关系都有点像那么回事。
我们其实能得出一个比较直接的判断——模型对空间结构的理解能力还是蛮到位的～
人家小哥也这么觉得。
他说他还拿这次GLM-5.1的模型和上一代模型进行了效果对比，确实觉得GLM-5.1的效果更好一些：
还有这个，他又尝试进行了游戏搭建，用GLM-5.1做了个可交互的国际象棋网页游戏，be like:
此外，他还用GLM-5.1搓出来了个在线版「我的世界」。
随着镜头的移动画面也没有出现卡顿的情况，而且模型对于未知画面的也能脑补出相应的画面。
在这个实测效果里，我们已经能浅浅推断出GLM-5.1模型在几个关键能力上的表现： 一是「空间一致性」。
场景、视角、交互是一起生成的，这说明模型在处理连续环境时，能呈现出一个相对稳定的空间状态。 二是「动态补全能力」。
随着镜头移动，对未出现区域的内容可以继续往下生成，也说明模型具备比较强大的持续构建环境的能力。
当然，除了这些实测和参数信息之外，我们从其他维度上也能对模型有更深的判断，那就是——「版本发布节奏」。
是的，这次GLM-5.1模型距离上一代GLM-5模型的发布，也就隔了一个多月的时间…
也就是说，这一代还处在一个非常快的迭代周期里。
所以说虽然版本不同，但两代在能力定位上是延续的，在GLM-5当时的公开资料里，就已经提到它在编程场景下的表现，已经逼近Claude Opus 4.5。
所以也不难得出的一个结论是—— GLM-5.1更像是上一代模型基础上的强化版本，这一代的提升，也会更多体现在稳定性、执行效率，以及复杂任务下的表现上。
问题来了，GLM-5.1咋具体配置呢，我们也帮大家浅浅整理了一下！
在配置层面，GLM-5.1目前已经支持接入Claude Code、OpenClaw等平台使用。
如果你是Mac用户，下面这套Claude Code配置GLM-5.1模型的流程可以直接照着操作～
* 方法一：在终端中运行vim ~/.claude/settings.json打开并编辑文件。完成后，按Esc，输入：wq，保存你的更改。 * 方法二：访达→前往文件夹→输入~/.claude/settings.json找到并编辑配置文件。
或者大家也可以直接用Claude Code代码来询问：
使用GLM-5.1，需要在配置文件settings.json中，添加或替换如下环境变量参数：
{
"env": {
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "GLM-4.5-air",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "GLM-5.1",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "GLM-5.1"
}
}
启动一个新的命令行窗口，运行claude启动Claude Code，在Claude Code中输入/status确认模型状态：
值得注意的是，官方说法是模型目前只能用于可以自定义模型的 Coding Agent。
若大家使用的Agent工具不能设置自定义模型，需要等待官方后续配置～
哦对了，Windows用户及OpenClaw的配置方法我也放到了下方链接里，友友们可以跟着教程进行操作。
（我还在苦苦候补订阅中，感兴趣的朋友可以直接上手试试～）
参考链接：
[1]模型入口：https://open.bigmodel.cn/glm-coding
[2]配置指南：https://docs.bigmodel.cn/cn/coding-plan/using5-1#glm-5-1
[3]https://www.youtube.com/watch?v=UxGieu7PaPg
[4]https://docs.z.ai/guides/llm/glm-5
量 量子位 @梦瑶
One Sentence Summary
Zhipu AI releases the GLM-5.1 model, featuring significantly improved programming capabilities that rival Claude Opus 4.6, with support for reasoning mode and integration with various programming tools.
Summary
This article reports on the surprise launch of Zhipu AI's GLM-5.1 model. This version sees a nearly 10-point increase in programming capability compared to its predecessor, with performance approaching the world-leading Claude Opus 4.6. The article details the model's key parameters (approx. 200K context, support for reasoning mode, OpenAI-compatible API), showcases real-world test results from users in scenarios like game development and spatial understanding, and provides a step-by-step guide for configuring GLM-5.1 in tools like Claude Code. Additionally, the article analyzes Zhipu's rapid iteration pace, noting that this version focuses on strengthening stability, execution efficiency, and performance in complex tasks.
Main Points
* 1. GLM-5.1 achieves a leap in programming capability, with performance rivaling industry benchmarks.
In official Coding Evaluation benchmarks, GLM-5.1 shows a nearly 10-point improvement over the previous generation, trailing the currently recognized top programming model, Claude Opus 4.6, by only 2.6 points.
* 2. The model's parameters and functional positioning are clear, focusing on complex code and reasoning scenarios.
It features a 200K context window, supports reasoning mode, and provides an OpenAI-compatible interface, significantly lowering the barrier to entry for developers.
* 3. Real-world testing demonstrates the model's evolution in spatial consistency and dynamic completion.
Through real-world test cases like an AI-powered 'Minecraft' and interior design blueprints, it showcases the model's advantages in handling continuous environments, spatial structure understanding, and long-text logic generation.
* 4. Provides highly valuable tool integration solutions.
It details the specific steps to integrate GLM-5.1 into Claude Code by modifying the settings.json configuration file, enabling developers to quickly incorporate it into their production workflows.
Key Quotes
* In terms of programming capability, it has surged by nearly 10 points compared to the previous generation GLM-5. * GLM-5.1 is more of an enhanced version based on the previous generation; the improvements in this generation are reflected more in stability, execution efficiency, and performance in complex tasks. * The barrier to entry has been pushed to the minimum; for developers, it means they can get the model running in their workflows without changing much code.
AI Score
87
Website qbitai.com
Published At Today
Length 2312 words (about 10 min)
Tags
GLM-5.1
Zhipu AI
Programming LLM
Claude Code
Model Evaluation
Related Articles
* GPT-5.4 Released: OpenAI's First Unified Model, Truly Native * Claude Code Founder's First Public Reveal: My 13 Usage Tips! * From ReAct to Ralph Loop: The Continuous Iteration Paradigm for AI Agents * Zhipu GLM-5 Technical Deep Dive: Full Compatibility with Domestic Chips Like Huawei Sparks Global Discussion and asynchronous reinforcement learning infrastructure, while completing full-stack adaptation for domestic chips.") * Google's New Paper Crashes Memory Stocks! 6x KV Cache Compression, Netizens: Silicon Valley Becomes Reality * AI Starts to "Take Action", Alibaba's Qwen Leads the World * From Clawdbot to the 2026 AI Coding Explosion | A Conversation with PingCAP CTO Dongxu * The Great Shift in Programming 2026: Anthropic Report Reveals Eight Major Trends in Agentic Programming * What You Don't Know About Claude Code: Architecture, Governance, and Engineering Practices - Tw93 * Yao Shunyu Lectures Face-to-Face with Tang Jie, Yang Zhilin, and Lin Junyang! Four Schema Heroes Debate Heroes at Zhongguancun HomeArticlesPodcastsVideosTweets