GLM-5.2 is the open-weight coding model from Z.ai (the international brand of Zhipu AI) that launched in June 2026. It was the strongest open model at the time, and it is still a strong choice. But it is no longer Zhipu AI's newest model. GLM-5.3 arrived in August, and that changes what you should pick today.
This guide covers the GLM-5.2 release date, what the model is, the costs, how to run it, and how it relates to GLM-5.3.
Key Importance:
GLM-5.2 was released on June 13, 2026, to Coding Plan users, with the API, chatbot, and open weights following on June 16.
It's a mixture-of-experts model, with a total number of parameters of approximately 750 billion, with around 40 billion parameters active per token and a context window of 1 million tokens.
The weights are free and available for download from Hugging Face and are licensed under MIT.
The API costs $1.40 per million input tokens and $4.40 per million output tokens.
The newer one is GLM-5.3 (August 14, 2026), but the weights are under a custom license rather than MIT.
What Is GLM-5.2?
GLM is the model series of Z.ai, a Chinese artificial intelligence (AI) company founded by researchers at Tsinghua University. Its flagship model and coding model is GLM-5.2, which is designed for long tasks in which a model needs to plan, edit code, test it, and correct its own errors over a number of steps.
It is defined by three aspects. It has three determining aspects:
These weights are under the MIT licence, so companies can freely download, fine-tune, and deploy commercially without the need for revenue or geographic restrictions.
A 1 million token context window (up from 200K for GLM-5.1) that provides a broad view of a large code base.
Mixture-of-experts architecture: inference cost is a fraction of the total size, as only a subset of the parameters are run for each token.
Sources estimate the total of 744-753 billion parameters, so round up to 750 billion.
GLM-5.2 Release Date and Model Timeline
The GLM-5 family moved fast in 2026:
February 2026: GLM-5 launches.
March 15: GLM-5-Turbo, a speed-focused API model.
April 1: GLM-5V-Turbo adds vision.
April 7: GLM-5.1 arrives as an open model.
June 13: GLM-5.2 goes out to GLM Coding Plan users
June 16: GLM-5.2 is injected into the standalone API, the chatbot and Hugging Face
August 14: GLM-5.3 is released via the API and Coding Plan
26th August: GLM-5.3: Flash lighter multimodal model is released under the MIT license
Looking for the GLM-5.2 release date? First access is June 13, 2026, and the general availability date is June 16, 2026. You may have seen June 16 listed as the primary release date on some sites, as this is when the public API and weights went live.
How Good Is GLM-5.2?
Z.ai did not publish a full benchmark table on launch day, so the best evidence is independent. Artificial Analysis scored GLM-5.2 at 51 on its Intelligence Index v4.1 at launch, the highest result for an open-weights model then. The US government's CAISI described it as probably the most capable open-weight model at the time of its release, with overall capability similar to GPT-5.2.
Two cautions apply. Benchmark scores depend on the index version, and Artificial Analysis has since moved to newer versions, so scores from different dates are not directly comparable. Also, many headline numbers for the GLM-5. X lines are reported by Z.ai itself. Run the model on your own tasks before you commit.
GLM-5.2 Pricing
On the Z.ai API, GLM-5.2 costs:
Input: $1.40 per million tokens
Cached input: $0.26 per million tokens
Output: $4.40 per million tokens
Third-party hosts such as OpenRouter sometimes list lower rates, so compare providers if volume is high. For subscriptions, the GLM Coding Plan has three tiers. Reported prices are $18, $80 and $168 per month on monthly billing for Lite, Pro and Max, or roughly $12.60, $56 and $117.60 per month on yearly billing. Check Z.ai's documentation for the live numbers, because plans and quotas have changed since launch.
The chatbot on the Z.ai site is free to use, and the open weights are free to download if you have the hardware.
How to Use the GLM-5.2 API
The model ID is glm-5.2, and requests go to Z.ai's API at https://api.z.ai/api/paas/v. 4. GLM-5.2 offers two reasoning modes. Therefore, you can trade depth for speed and cost. The Coding Plan also works inside popular coding tools such as Claude Code, which makes switching straightforward if you already use one of them.
One practical warning: if you call GLM-5.2 through the Coding Plan, Z.ai now routes the request to GLM-5.3. On the pay-as-you-go API, both models currently list at the same price. If you need results that stay identical over time, pin your model ID and retest whenever changes routing.
Running GLM-5.2 Locally
The weights are on Hugging Face as zai-org/GLM-5.2, with an FP8 build alongside it. They run through SGLang, vLLM, Transformers, KTransformers and Unsloth, and quantised GGUF versions exist for llama.cpp, Ollama and LM Studio.
Be realistic about hardware. The full model is over 1.5 TB, so this is a multi-GPU job, not a laptop one. For smaller setups, our guide to the Qwen3-Coder and Ollama stack is a better fit, and a hosted provider is the practical route for most teams.
GLM-5.2 vs GLM-5.3
GLM-5.3 is the same 753-billion-parameter base with improvements made through further post-training, aimed at coding, agents and security tasks. It costs the same on the API, but its open weights, published August 25, use a custom GLM-5.3 license instead of MIT. That is the main reason to stay on GLM-5.2: if your legal team requires an approved license such as MIT, or you plan to redistribute weights, GLM-5.2 is the cleaner choice.
For everyone else, GLM-5.3 is the better default. Z.ai's own figures claim large coding gains over GLM-5.2, though these are vendor numbers. If you want a smaller option, GLM-5.3-Flash is natively multimodal and MIT licensed, whereas GLM-5.2 is not, since GLM-5.2 handles text only.
How GLM-5.2 Compares With Other Open Models
GLM-5.2 competes with other Chinese open-weight models, and the field has moved quickly. Our guides to DeepSeek V4 and Kimi K3 cover the main alternatives. On price, GLM-5.2's output rate of $4.40 sits far below closed models such as GPT-5.5, which we cover in our GPT-5.5 overview. If you are weighing open weights against hosted APIs, check the Artificial Analysis page for GLM-5.2 for independent comparisons.
Who Should Use GLM-5.2?
It suits teams that run coding agents, work with very large codebases, or need self-hosted AI under a permissive license. It suits less well teams that need image input, a fully managed vendor relationship, or the newest model by default. In that case, look at GLM-5.3 or a closed model.
Conclusion
GLM-5.2 was a turning point for open models, and it remains a good pick if you want MIT-licensed weights and a 1 million token context. For new projects, check GLM-5.3 first, then decide whether the license difference matters to you. Whatever you choose, verify prices and model names in the official docs, because both have moved several times this year.
Need help choosing and deploying the right AI model for your team? Talk to TechNow, and we will help you compare, test, and set up the right solution.
Frequently Asked Questions
What is GLM-5.2?
Z.ai's open-weights coding and agent model, with a 1 million token context window and an MIT license.
When was GLM-5.2 released?
June 13, 2026, for Coding Plan users, and June 16, 2026, for the API, chatbot, and open weights.
Who makes GLM-5.2?
Z.ai, also known as Zhipu AI, is a Chinese company that came out of Tsinghua University.
How much does GLM-5.2 cost?
$1.40 per million input tokens, $0.26 per million cached tokens, and $4.40 per million output tokens on the Z.ai API.
Is GLM-5.2 open source?
The weights are MIT-licensed on Hugging Face, so you can download and use them commercially.
Is there a newer model than GLM-5.2?
Yes. GLM-5.3 launched in August 2026, and GLM-5.3-Flash followed on August 26.
Can GLM-5.2 process images?
No. GLM-5.2 is text-only. GLM-5.3-Flash is the multimodal option.