KAT Coder V2.5 Dev needs about 22.6 GB of memory at Q4_K_M (a 21.4 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 28 quants, from 0.2 GB to 69.4 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.7 GB.
35B‑parameter MOE coding model with 8‑bit context
KAT Coder V2.5 Dev is a 35 B parameter mixture‑of‑experts model built on the qwen35moe architecture. It is trained for coding tasks and supports up to 262 144 context tokens. The model is released under the Apache‑2.0 license and is text‑only.
Good at
- High‑quality code generation
- Large context handling
- Fine‑tuned agentic coding
Limits
- No vision or multimodal support
- Requires large VRAM for higher quants
- License allows open use but not redistribution of weights
Best for
- Writing complex scripts
- Debugging and refactoring code
- Automated coding agents
Who it’s for. Developers and researchers who need a large‑scale coding model that can handle long contexts. The model is suitable for integration into tools that support llama.cpp or similar inference engines.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.2 GB | 1.4 GB | Fits | Get file → | |
| IQ2_XXS | 9.8 GB | 10.9 GB | Too big | Get file → | |
| IQ2_XS | 10.8 GB | 12.0 GB | Too big | Get file → | |
| IQ2_S | 11.0 GB | 12.2 GB | Too big | Get file → | |
| IQ2_M | 12.1 GB | 13.2 GB | Too big | Get file → | |
| Q2_K | 12.6 GB | 13.8 GB | Too big | Get file → | |
| Q2_K_L | 13.1 GB | 14.3 GB | Too big | Get file → | |
| IQ3_XXS | 14.9 GB | 16.0 GB | Too big | Get file → | |
| Q3_K_S | 15.5 GB | 16.7 GB | Too big | Get file → | |
| IQ3_XS | 16.2 GB | 17.4 GB | Too big | Get file → | |
| Q3_K_M | 16.2 GB | 17.4 GB | Too big | Get file → | |
| Q3_K_L | 16.9 GB | 18.1 GB | Too big | Get file → | |
| IQ3_M | 16.9 GB | 18.1 GB | Too big | Get file → | |
| Q3_K_XL | 17.3 GB | 18.5 GB | Too big | Get file → | |
| IQ4_XS | 18.8 GB | 20.0 GB | Too big | Get file → | |
| IQ4_NL | 19.9 GB | 21.0 GB | Too big | Get file → | |
| Q4_0 | 19.9 GB | 21.1 GB | Too big | Get file → | |
| Q4_K_S | 20.6 GB | 21.8 GB | Too big | Get file → | |
| Q4_K_M | 21.4 GB | 22.6 GB | Too big | Get file → | |
| Q4_K_L | 21.8 GB | 22.9 GB | Too big | Get file → | |
| Q4_1 | 22.0 GB | 23.1 GB | Too big | Get file → | |
| Q5_K_S | 24.2 GB | 25.3 GB | Too big | Get file → | |
| Q5_K_M | 25.0 GB | 26.2 GB | Too big | Get file → | |
| Q5_K_L | 25.3 GB | 26.5 GB | Too big | Get file → | |
| Q6_K | 30.1 GB | 31.2 GB | Too big | Get file → | |
| Q6_K_L | 30.3 GB | 31.5 GB | Too big | Get file → | |
| Q8_0 | 36.9 GB | 38.1 GB | Too big | Get file → | |
| BF16 | 69.4 GB (2 parts) | 70.5 GB | Too big | Get file → |
For 8 GB VRAM, use IQ2_XXS (0.19 GB file, ~1.4 GB VRAM). For 16 GB VRAM, IQ4_XS (18.81 GB file, ~20 GB VRAM) or Q4_K_S (20.59 GB file, ~21.8 GB VRAM). For 24 GB VRAM, Q5_K_M (25.02 GB file, ~26.2 GB VRAM).
What GPU do you need for KAT Coder V2.5 Dev?
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: the best quant that fits is the main file (0.2 GB file, about 1.4 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (0.2 GB file, about 1.4 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (0.2 GB file, about 1.4 GB in use).
- 12 GB VRAM: the best quant that fits is IQ2_XS (10.8 GB file, about 12.0 GB in use).
- 16 GB VRAM: the best quant that fits is IQ3_XXS (14.9 GB file, about 16.0 GB in use).
- 24 GB VRAM: the best quant that fits is Q4_1 (22.0 GB file, about 23.1 GB in use).
- 32 GB VRAM: the best quant that fits is Q6_K_L (30.3 GB file, about 31.5 GB in use).
- 48 GB VRAM: the best quant that fits is Q8_0 (36.9 GB file, about 38.1 GB in use).
Which KAT Coder V2.5 Dev quant should you choose?
Q4_K_M is the usual starting point for KAT Coder V2.5 Dev: it balances file size and quality well. Here it is a 21.4 GB file.
The Q8_0 file is 36.9 GB, about 1.7 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 the first one in the table at 0.2 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about KAT Coder V2.5 Dev
How much VRAM does KAT Coder V2.5 Dev need?
About 22.6 GB at Q4_K_M with an 8K context window. That is the 21.4 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 KAT Coder V2.5 Dev on an 8 GB GPU?
Yes, with the smaller files: the 0.2 GB file needs about 1.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of KAT Coder V2.5 Dev GGUF?
It depends on the quant: file is 0.2 GB, IQ2_XXS is 9.8 GB, IQ2_XS is 10.8 GB, IQ2_S is 11.0 GB, IQ2_M is 12.1 GB, Q2_K is 12.6 GB and 22 more.
What context length does KAT Coder V2.5 Dev support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.7 GB at 8K and 2.7 GB at 32K.
Where can I download KAT Coder V2.5 Dev GGUF?
The files are on Hugging Face in the bartowski/Kwaipilot_KAT-Coder-V2.5-Dev-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 KAT Coder V2.5 Dev use?
Hugging Face lists the license as apache-2.0. 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 KAT Coder V2.5 Dev on 8 GB VRAM?
Yes, using the IQ2_XXS quant which requires about 1.4 GB VRAM at 8k context.
Does the model support image input?
No, the released weights are text‑only; vision components are not included.
What license applies to this model?
The model is released under the Apache‑2.0 license, allowing free use and modification.