cerebras/GLM-4.6-REAP-268B-A32B needs about 166.1 GB of memory at Q4_K_M (a 162.5 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 27 quants, from 64.8 GB (TQ1_0) to 537.7 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 12.3 GB.
GLM-4.6-REAP-268B-A32B – a 268B sparse MoE model compressed 25% from 355B
GLM-4.6-REAP-268B-A32B is a memory‑efficient compressed variant of GLM‑4.6. It retains near‑lossless performance while reducing parameters to 268B and activating 32B per token. The model supports a context length of 202,752 tokens and is licensed under MIT.
Good at
- Reasoning tasks
- Tool usage
- Code generation
Limits
- No vision support
- Requires large VRAM for full precision
- Context limited to 202,752 tokens
Best for
- Local deployment with limited GPU memory
- Academic research on sparse MoE models
- Fine‑tuning on domain‑specific data
Who it’s for. Researchers and developers needing a high‑capacity language model that fits within constrained GPU resources. The model is suitable for environments where memory efficiency is critical, such as edge devices or small‑scale servers.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| TQ1_0 | 64.8 GB | 68.4 GB | Too big | Get file → | |
| IQ1_S | 74.2 GB (2 parts) | 77.8 GB | Too big | Get file → | |
| IQ1_M | 82.4 GB (2 parts) | 86.0 GB | Too big | Get file → | |
| IQ2_XXS | 89.1 GB (2 parts) | 92.7 GB | Too big | Get file → | |
| IQ2_M | 93.7 GB (2 parts) | 97.3 GB | Too big | Get file → | |
| Q2_K | 98.7 GB (2 parts) | 102.3 GB | Too big | Get file → | |
| Q2_K_L | 98.9 GB (2 parts) | 102.5 GB | Too big | Get file → | |
| Q2_K_XL | 103.0 GB (3 parts) | 106.6 GB | Too big | Get file → | |
| IQ3_XXS | 110.3 GB (3 parts) | 113.9 GB | Too big | Get file → | |
| Q3_K_S | 116.5 GB (3 parts) | 120.1 GB | Too big | Get file → | |
| Q3_K_XL | 120.3 GB (3 parts) | 123.9 GB | Too big | Get file → | |
| Q3_K_M | 128.7 GB (3 parts) | 132.3 GB | Too big | Get file → | |
| IQ4_XS | 143.7 GB (3 parts) | 147.3 GB | Too big | Get file → | |
| IQ4_NL | 152.0 GB (4 parts) | 155.6 GB | Too big | Get file → | |
| Q4_0 | 152.4 GB (4 parts) | 156.0 GB | Too big | Get file → | |
| Q4_K_S | 152.9 GB (4 parts) | 156.5 GB | Too big | Get file → | |
| Q4_K_XL | 153.9 GB (4 parts) | 157.4 GB | Too big | Get file → | |
| Q4_K_M | 162.5 GB (4 parts) | 166.1 GB | Too big | Get file → | |
| Q4_1 | 168.6 GB (4 parts) | 172.2 GB | Too big | Get file → | |
| Q5_K_S | 185.4 GB (4 parts) | 189.0 GB | Too big | Get file → | |
| Q5_K_M | 190.8 GB (4 parts) | 194.4 GB | Too big | Get file → | |
| Q5_K_XL | 190.9 GB (4 parts) | 194.5 GB | Too big | Get file → | |
| Q6_K | 220.9 GB (5 parts) | 224.4 GB | Too big | Get file → | |
| Q6_K_XL | 231.1 GB (5 parts) | 234.7 GB | Too big | Get file → | |
| Q8_0 | 285.8 GB (6 parts) | 289.4 GB | Too big | Get file → | |
| Q8_K_XL | 310.6 GB (7 parts) | 314.1 GB | Too big | Get file → | |
| BF16 | 537.7 GB (11 parts) | 541.3 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_0 or Q4_K_S; they require approximately 156 GB of VRAM at 8 k context, which is unrealistic, so a lower‑bit quant like Q4_0 is recommended for smaller GPUs. For 16 GB VRAM, Q4_K_M or Q4_1 provide a balance between size and performance. For 32 GB VRAM, Q5_K_S or Q5_K_M offer higher quality while staying within memory limits.
What GPU do you need for cerebras/GLM-4.6-REAP-268B-A32B?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: no quant fits at an 8K context.
- 32 GB VRAM: no quant fits at an 8K context.
- 48 GB VRAM: no quant fits at an 8K context.
Which cerebras/GLM-4.6-REAP-268B-A32B quant should you choose?
Q4_K_M is the usual starting point for cerebras/GLM-4.6-REAP-268B-A32B: it balances file size and quality well. Here it is a 162.5 GB file.
The Q8_0 file is 285.8 GB, about 1.8 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is TQ1_0 at 64.8 GB, with the quality trade-offs that come with fewer bits. Lower bit counts mean smaller files and more quality loss.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about cerebras/GLM-4.6-REAP-268B-A32B
How much VRAM does cerebras/GLM-4.6-REAP-268B-A32B need?
About 166.1 GB at Q4_K_M with an 8K context window. That is the 162.5 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run cerebras/GLM-4.6-REAP-268B-A32B on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 68.4 GB, so you would need a larger GPU or CPU offloading.
What is the file size of cerebras/GLM-4.6-REAP-268B-A32B GGUF?
It depends on the quant: TQ1_0 is 64.8 GB, IQ1_S is 74.2 GB, IQ1_M is 82.4 GB, IQ2_XXS is 89.1 GB, IQ2_M is 93.7 GB, Q2_K is 98.7 GB and 21 more.
What context length does cerebras/GLM-4.6-REAP-268B-A32B support?
The GGUF header lists a trained context length of 202,752 tokens (198K). The KV cache grows with the context you use: about 3.1 GB at 8K and 12.3 GB at 32K.
Where can I download cerebras/GLM-4.6-REAP-268B-A32B GGUF?
The files are on Hugging Face in the unsloth/GLM-4.6-REAP-268B-A32B-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does cerebras/GLM-4.6-REAP-268B-A32B use?
Hugging Face lists the license as mit. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run GLM-4.6-REAP-268B-A32B on 8 GB VRAM?
Running the full model on 8 GB VRAM is not feasible. Use a low‑bit quant such as Q4_0 or Q4_K_S to reduce VRAM usage, but performance will be lower.
Does the model support vision tasks?
No, the model does not include vision capabilities.
What license does the model use?
The model is released under the MIT license.