GLM-5.2
Z.ai's MIT-licensed 753B flagship with a 1M-token context and no regional limits; GLM-5.3 is its newer successor.
In short. GLM-5.2 is a very large text model from Z.ai for coding and long, multi-step tasks. Its license is plain MIT, so companies can use and change it freely. It needs a GPU server or a machine with very large memory, and Z.ai has since released the newer GLM-5.3.
Summary
Z.ai released GLM-5.2 on 16 June 2026 as its flagship for long-horizon tasks, after GLM-5.1 in April. It is a text-only mixture-of-experts model with about 753 billion parameters. The configuration lists 78 layers and 256 routed experts, with 8 active per token plus one shared expert, and DeepSeek-style sparse attention. The configuration is identical to GLM-5.3's, which has about 40 billion active parameters, and the model card does not state the figure itself. The context window is 1,048,576 tokens.
Z.ai highlights three things. First, a "solid" 1M-token context that holds up on long tasks. Second, an index-sharing scheme the card calls IndexShare, which reuses one sparse-attention indexer across every four layers and, per the card, cuts per-token compute by 2.9 times at 1M context. Third, an improved multi-token-prediction layer for speculative decoding with up to 20 percent longer accepted drafts. Several thinking-effort levels let you trade quality for latency.
Z.ai's table reports SWE-Bench Pro 62.1, Terminal-Bench 2.1 81.0 with Terminus-2, MCP-Atlas 76.8 and HLE 40.5 on the text-only subset, set against GLM-5.1 and models from Anthropic, OpenAI, Google, Alibaba, MiniMax and DeepSeek. All of it is self-reported. The BF16 repository is about 1.5 TB and an FP8 build of about 756 GB is published separately. It runs on SGLang, vLLM, Transformers, KTransformers and Unsloth, and on Ascend NPU stacks. The license is MIT, unlike the custom GLM-5.3 License with its security-review clause for very large hosting businesses.
Good for
- Self-hosted coding agents and long-horizon engineering tasks
- Long-context work up to 1 million tokens
- Organizations that want a plain MIT license with no regional or revenue conditions
- Research on sparse attention and large mixture-of-experts serving
Watch out for
- GLM-5.3 is newer and Z.ai's card points to it, so check whether you need the license difference before choosing 5.2
- It needs a multi-GPU node even in FP8, or a machine with roughly 512 GB of fast memory for a 4-bit build
- It is text-only; GLM-5.3-Flash accepts images
- Benchmark numbers come from Z.ai and use its own harness settings
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Estimates for the GLM-5.2 (BF16) variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.
| Variants | Parameters | BF16 | FP8 / INT8 | 4-bit (Q4) | Fits on |
|---|---|---|---|---|---|
| GLM-5.2 (BF16) BF16 | 753B 40B active | 1.73 TB | 866 GB | 520 GB | GPU server |
| GLM-5.2-FP8 FP8 | 753B 40B active | 1.73 TB | 866 GB | 520 GB | GPU server |
Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.
The license in plain words
The shortest and simplest permissive license. You can do almost anything with the model as long as you keep the copyright and license notice.
Frequently asked questions
- What is GLM-5.2?
- GLM-5.2 is a very large text model from Z.ai for coding and long, multi-step tasks. Its license is plain MIT, so companies can use and change it freely. It needs a GPU server or a machine with very large memory, and Z.ai has since released the newer GLM-5.3.
- Can I use GLM-5.2 commercially?
- Yes. GLM-5.2 is released under MIT, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
- Is GLM-5.2 open source?
- GLM-5.2 is released under MIT, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
- Can I run GLM-5.2 on my own computer?
- Not on a desktop. The smallest variant, GLM-5.2 (BF16), needs about 520 GB at 4-bit precision, which means a GPU server with several data-center cards.
- When was GLM-5.2 released?
- GLM-5.2 was released by Z.ai (Zhipu AI) on June 16, 2026.
- How long a context window does GLM-5.2 have?
- GLM-5.2 supports a context window of up to 1M tokens.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- GLM-5.2 model card (Hugging Face) huggingface.co
- Z.ai release notes (GLM-5.2, 16 Jun 2026) docs.z.ai
- GLM-5: from Vibe Coding to Agentic Engineering (technical report) arxiv.org
- IndexCache: Accelerating Sparse Attention via Cross-Layer Index Reuse arxiv.org
- GLM-5 repository (GitHub) github.com
Release date and facts last verified November 11, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| GLM-5.3 Z.ai (Zhipu AI) | 753B 40B active | 1M | GLM-5.3 License (Conditional) | 520 GB GPU server | |
| GLM-5.3-Flash Z.ai (Zhipu AI) | 320B 18B active | 1M | MIT (Permissive) | 221 GB Big-memory desktop | |
| GLM-5.1 Z.ai (Zhipu AI) | 754B 40B active | 198K | MIT (Permissive) | 520 GB GPU server | |
| DeepSeek V4 DeepSeek | 284B – 1.6T 49B active | 1M | MIT (Permissive) | 176 GB Big-memory desktop |