GLM 5.2 is out: The Code Model That Reads Your Entire Project (1M Tokens)

Original author：Matheus Battisti - Hora de Codar

Original source：https://www.youtube.com/embed/yhmeF_AFZpM

⭐ Sign up for the Stop Fighting with AI Course Waiting List: https://app.horadecodar.com.br/lp/pare-de-brigar-com-a-ia-claude-code 🔴 Vibe Coding Training (Antigravity, Claude Code and +): https://app.horadecodar.com.br/lp/formacao-vibe-coding?utm_source=yt_glm_52 🟪 Hosting that I recommend: https://www.hostinger.com/matheushoradecodar (use the coupon HORADECODAR to get +10% discount) 📘 Guia Engenharia de Prompt: https://app.horadecodar.com.br/ebookpages/guia-engenharia-de-prompt Join our Discord server and follow me on the networks: 🟣 Discord Hora de Codar: https://discord.gg/Veq4mvsWwk 🔴 Instagram: https://www.instagram.com/horadecodar/ GLM 5.2 is the new Zhipu AI (Z.ai) code model, released on June 13, 2026, focusing on programming and long agentic tasks. The main highlight is the 1 million token context window, five times larger than that of GLM 5.1, which allows you to work with entire large repositories in a single session. In this video I show what it is, the real specs, how to run it on Claude Code, and I give my evaluation without inflating: it is a competent model, but it has important caveats that are worth mentioning. If you follow open source code models and want to understand where GLM 5.2 fits in before investing time in it, this video is for you. It joins the line of Chinese models focused on code that we have been following, and Zhipu's proposal is clear: to offer a cheaper alternative to closed models, positioning the GLM Coding Plan at around a tenth of the price of Claude Code's premium plans. In specs, GLM 5.2 is a Mixture of Experts model with 744 billion parameters in total and around 40 billion active at a time. It has a context of 1 million input tokens and up to 131 thousand output tokens, which allows for refactoring on a repository scale and long planning and execution traces. It has two reasoning modes, High and Max, so you can balance latency and quality depending on the task. The license is MIT, so the open weights will be able to be lowered and rotated freely, including commercially. An important practical point: GLM 5.2 exposes an endpoint compatible with the Anthropic format, that is, you can plug it directly into Claude Code, Cline, OpenCode and similar tools, without changing environments. Just point the configuration to the Z.ai endpoint. In the video I show this setup. About access and price: at launch, GLM 5.2 was available in all tiers of the GLM Coding Plan (Lite, Pro, Max and Team), with the entry tier costing around US$10 per month. The standalone API, the Z.ai chatbot and the open weights under the MIT license were announced for the week after the launch, so some access was still coming when the model came out. Now the part that needs to be said clearly: Zhipu did not publish any official benchmarks at launch. There is no SWE-Bench, there is no LiveCodeBench, there is no verified number. All that exists so far are marketing claims from the company itself, such as "powerful code" and "strong in long tasks", with no data to support it. So any direct comparison with Claude, GPT or other models, at this point, is speculation. This is exactly why I tested it in practice instead of repeating promises. And my honest assessment is this: GLM 5.2 is no savior. The context of 1 million tokens is real and useful for those who work with a large codebase, and the low price is a legitimate attraction for those who want to cut costs. But in real use it didn't impress me enough to recommend changing what you already use. It's a cheap option with generous context, not a frontier model. As always, the best thing to do is test it in your case. It's worth noting the context of the launch, which was curious: GLM 5.2 came out shortly after Anthropic suspended access to the Fable 5 and Mythos 5 models outside the United States due to an export control directive. Zhipu took advantage of the moment to reinforce the discourse of open and accessible intelligence, and the company's shares rose sharply on the stock market. It's another chapter in the dispute between Western closed models and Chinese open sources, and GLM 5.2 is the most recent piece on that board. And you, have you tested the GLM 5.2 or any other model from the GLM line? Comment below what you think, and which open source model you want to see tested on the channel next. Leave a like, subscribe to the channel and activate the bell so you don't miss the next videos about open source models, Claude Code and AI tools. LINKS GLM 5.2 (Z.ai platform): https://z.ai TIMESTAMPS 00:00 You released GLM 5.2, is it good? 00:55 How to use GLM 5.2 in Claude Code 04:11 Creating PRD with GLM 5.2 for the project 07:20 Starting the project with GLM 5.2 in Next JS 08:40 Testing the project created with GLM 5.2 09:30 Implementing project authentication with GLM 5.2 12:00 Integrating AI into the project 17:05 Benchmarks and analysis about GLM 5.2 19:05 Final considerations about GLM 5.2