Run the new GLM-5.2 model by Z.ai on local hardware!
GLM-5.2 is Z.ai’s new open model, delivering SOTA performance across long-horizon coding, reasoning, and agentic tasks. With 744B parameters, 40B active parameters, and a 1M context window, it can now be run locally using Unsloth Dynamic GGUFs. GLM-5.2 is the strongest open model to date, performing on par with Claude 4.8 Opus, GPT-5.5, and Gemini 3.1 Pro across Artificial Analysis and many other benchmarks.
The full model requires 1.51TB of disk space, while Unsloth Dynamic 2-bit GGUF reduces this to **239GB (-84% size) by **upcasting important layers to 8 or 16-bit. Dynamic 1-bit lowers further to 217GB (-86%). Thanks Z.ai for giving Unsloth day-zero access. GLM-5.2-GGUF
Run GLM-5.2 TutorialsQuantization Results ** **⚙️ Usage Guide
⚙️ Usage Guide
The 2-bit dynamic quant UD-IQ2_M
uses 239GB of disk space - this can directly fit on a **256GB unified memory Mac **and works well in a 1x24GB GPU and 256GB of RAM with MoE off. The 1-bit quant will fit on a 223GB RAM and 8-bit requires 810GB RAM.
Table: Inference hardware requirements (units = total memory: RAM + VRAM, or unified memory) 223 GB
245 GB
290-360 GB
372-475 GB
570 GB
810 GB
For best performance, make sure your total available memory, including VRAM and system RAM, exceeds the quantized model file size by a comfortable margin.
Recommended Settings
GLM-5.2 has 3 thinking modes. Non-thinking and Thinking in two modes: High + Max. Use Max Thinking for complicated tasks. In Unsloth Studio you can easily toggle High + Max Thinking and non-Thinking with a UI.
Use these settings for most use cases:
temperature
= 1.0
temperature
= 1.0
top_p
= 0.95
top_p
= 1.0
Maximum context window:1,048,576
.
GLM 5.2 uses thinking mode by default. And supports reasoning_effort
as "high", "max" or disabled thinking. To disable thinking, use `--chat-template-kwargs '{"enable_thinking":false}'`
If you're on **Windows** Powershell, use: `--chat-template-kwargs "{\"enable_thinking\":false}"`
Use 'true' and 'false' interchangeably.
You can also use `--reasoning on`
or `--reasoning off`
in llama.cpp as well now!
📈 Quantization analysis
We also ran KLD (KL Divergence) to gauge the accuracy of our quantizations of GLM-5.2-GGUF. In general, dynamic 4-bit UD-Q4_K_XL and dynamic 5-bit UD-Q5_K_XL are generally lossless, and smaller quants also work great!
On pure top-1% accuracy, dynamic 1-bit gets around 76.2% accuracy yet being 86% smaller! Dynamic 2-bit gets around 82% accuracy whilst being 84% smaller.
99.9% KLD is also generally good - there is a larger uplift from 4bit onwards though, so for massive out of distribution tasks, dynamic 4-bit is probably best.
The mean KLD generally follows a clear monotonic trend vs disk space, and shows even at 1-bit GLM 5.2 works well!
Run GLM-5.2 Tutorials: #
You can now run GLM-5.2 in llama.cpp and Unsloth Studio. We will be utilizing the 239GB UD-IQ2_M quant for best results in terms of accessbility and accuracy.
🦥 Run GLM-5.2 in Unsloth Studio
GLM-5.2 can run in Unsloth Studio, an open-source web UI for local AI. Unsloth Studio automatically offloads to RAM and detects multiGPU setups. With Unsloth Studio, you can run models locally on MacOS, Windows, Linux and:
Search, download,
run GGUFsand safetensor models+Self-healing tool callingweb search****Code execution(Python, Bash)Automatic inferenceparameter tuning (temp, top-p, etc.)Fast CPU + GPU inference via llama.cpp
Train LLMs2x faster with 70% less VRAM Install and Launch Unsloth
To install, run in your terminal:
MacOS, Linux, WSL:
Windows PowerShell:
Launch Unsloth
MacOS, Linux, WSL and Windows:
Then open `http://127.0.0.1:8888`
(or your specific URL) in your browser.
Launch Unsloth securely with HTTPS and Cloudflare
**NEW! **Unsloth now provides a secure way to launch Studio over HTTPS through a free Cloudflare tunnel. Use the below (works in Windows, Mac & Linux):
Search and download GLM-5.2
Unsloth Studio automatically offloads to RAM and detects multiGPU setups. On first launch you will need to create a password to secure your account and sign in again later.
Then go to the Studio Chat tab and search for GLM-5.2 in the search bar and download your desired model and quant. Ensure you have enough compute the run the model.
Run GLM-5.2
Inference parameters should be auto-set when using Unsloth Studio, however you can still change it manually. You can also edit the context length, chat template and other settings.
For more information, you can view our Unsloth Studio inference guide.
🦙 Run GLM-5.2 in llama.cpp
For this guide we'll be running the UD-IQ2_M
quant which will require at least 245GB RAM. Feel free to change quantization type. For these tutorials, we will using llama.cpp for fast local inference. GGUF: GLM-5.2-GGUF** **
Obtain the latest llama.cpp
on GitHub here. You can follow the build instructions below as well. Change
`-DGGML_CUDA=ON`
to `-DGGML_CUDA=OFF`
if you don't have a GPU or just want CPU inference. **For Apple Mac / Metal devices**, set
`-DGGML_CUDA=OFF`
then continue as usual - Metal support is on by default.You can now use llama.cpp
directly to load and download models, just like ollama run
. First, select the quantization type you want like UD-IQ2_M
. Also use export LLAMA_CACHE="unsloth/GLM-5.2-GGUF"
to force llama.cpp
to save to a specific location. Note this download process might be very slow, so it's probably best to use the manual download process in the next section.
If you want to download the model manually **(much faster!)**, we can download the model via the code below (after installing `pip install huggingface_hub`
). If downloads get stuck, see: [Hugging Face Hub, XET debugging](/docs/basics/troubleshooting-and-faqs/hugging-face-hub-xet-debugging)
If you want to use the dynamic 1bit, then do:
Then run the model in conversation mode. Use `unsloth/GLM-5.2-GGUF/UD-IQ2_M/GLM-5.2-UD-IQ2_M-00001-of-00006.gguf`
for 2bit or `unsloth/GLM-5.2-GGUF/UD-IQ1_S/GLM-5.2-UD-IQ1_S-00001-of-00006.gguf`
for 1bit.
📐Long context via KV Cache quantization
To utilize long context in llama.cpp, we need to employ KV cache quantization to reduce memory usage. Recently llama.cpp added higher accuracy tricks to KV cache quantization - see and other PRs!
Currently, these KV cache dtypes are supported:
By default f16
is used. If you use q4_0
which is around 4.5 bits per weight, you can extend around 16 / 4.5 = 3.5x longer context lengths! So if you model used to support 10K, 35K can be in reach! q4_1
is probably better since you also get a shifting parameter, and is 5 bits per weight - so 3.2x longer contexts.
Use it like below:
📊 Benchmarks #
You can view further below for GLM-5.2 benchmarks in table format:
Reasoning
HLE
40.5
49.8*
41.4*
45
31
41.4
37
37.7
HLE (w/ Tools) 54.7
57.9*
52.2*
51.4*
52.3
53.5
48.2
CritPt
20.9
20.9
27.1
17.7
4.6
13.4
3.7
12.9
AIME 2026
99.2
95.7
98.3
98.2
95.3
97
94.6
HMMT Nov. 2025
94.4
96.5
96.5
94.8
94
95
84.4
94.4
HMMT Feb. 2026
92.5
96.7
96.7
87.3
82.6
97.1
84.4
95.2
IMOAnswerBench
91.0
83.5
81
83.8
90
89.8
GPQA-Diamond
91.2
93.6
93.6
94.3
86.2
90
93
90.1
Coding
SWE-bench Pro
62.1
69.2
58.6
54.2
58.4
60.6
59
55.4
NL2Repo
48.9
69.7
50.7
33.4
42.7
47.2
42.1
35.5
DeepSWE
46.2
58
70
10
18
18
20
8
ProgramBench
63.7
71.9
70.8
39.5
50.9
47.8
Terminal Bench 2.1 (Terminus-2) 81.0
85
84
74
63.5
75
65
64
Terminal Bench 2.1 (Best Reported Harness)
82.7
78.9
83.4
70.7
69
FrontierSWE (Dominance) 74.4
75.1
72.6
39.6
30.5
29.0
PostTrainBench
34.3
37.2
28.4
21.6
20.1
SWE-Marathon
13.0
26.0
12.0
4.0
1.0
Agentic
MCP-Atlas (Public Set) 76.8
77.8
75.3
69.2
71.8
76.4
74.2
73.6
Tool-Decathlon
48.2
59.9
55.6
48.8
40.7
52.8
Last updated
Was this helpful?