Kimi K3 0.40B needs about 1.1 GB of memory at Q4_K_M (a 0.3 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 25 quants, from 0.0 GB to 0.3 GB (Q6_K).
Kimi K3 0.40B – a small, mixed‑expert language model
Kimi K3 0.40B is a 0.40B parameter model built on the kimi-k3 architecture. It uses a 4096 token context window and a sparse MoE with two experts active per token. The model is released under the MIT license.
Good at
- Small footprint for local inference
- Supports 4096 token context
- Mixture of experts for efficient scaling
Limits
- No vision training support
- Requires custom KimiK3ForConditionalGeneration code
- Limited to 4096 token context
Best for
- Rapid prototyping on low‑VRAM GPUs
- Testing new quantization schemes
- Educational demos of MoE models
Who it’s for. Developers and researchers who need a lightweight Kimi‑K3 variant for local testing. The model is ideal for those who want to experiment with custom quantization or evaluate MoE behavior without large GPU resources.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.0 GB | 0.7 GB | Fits | Get file → | |
| IQ1_S | 0.2 GB | 0.8 GB | Fits | Get file → | |
| IQ1_M | 0.2 GB | 0.8 GB | Fits | Get file → | |
| IQ2_XXS | 0.2 GB | 0.8 GB | Fits | Get file → | |
| IQ2_XS | 0.2 GB | 0.8 GB | Fits | Get file → | |
| IQ2_S | 0.2 GB | 0.9 GB | Fits | Get file → | |
| IQ2_M | 0.2 GB | 0.9 GB | Fits | Get file → | |
| IQ3_XXS | 0.2 GB | 0.9 GB | Fits | Get file → | |
| Q2_K_S | 0.2 GB | 0.9 GB | Fits | Get file → | |
| Q2_K | 0.2 GB | 0.9 GB | Fits | Get file → | |
| IQ3_XS | 0.2 GB | 0.9 GB | Fits | Get file → | |
| IQ3_S | 0.2 GB | 0.9 GB | Fits | Get file → | |
| Q3_K_S | 0.2 GB | 0.9 GB | Fits | Get file → | |
| IQ3_M | 0.2 GB | 0.9 GB | Fits | Get file → | |
| Q3_K_M | 0.2 GB | 0.9 GB | Fits | Get file → | |
| Q3_K_L | 0.2 GB | 0.9 GB | Fits | Get file → | |
| IQ4_XS | 0.3 GB | 0.9 GB | Fits | Get file → | |
| Q4_0 | 0.3 GB | 0.9 GB | Fits | Get file → | |
| IQ4_NL | 0.3 GB | 0.9 GB | Fits | Get file → | |
| Q4_K_S | 0.3 GB | 0.9 GB | Fits | Get file → | |
| Q4_K_M | 0.3 GB | 0.9 GB | Fits | Get file → | |
| Q4_1 | 0.3 GB | 0.9 GB | Fits | Get file → | |
| Q5_K_S | 0.3 GB | 0.9 GB | Fits | Get file → | |
| Q5_K_M | 0.3 GB | 0.9 GB | Fits | Get file → | |
| Q6_K | 0.3 GB | 1.0 GB | Fits | Get file → |
For 8 GB VRAM, use IQ1_S or IQ1_M (0.19 GB file, ~1 GB memory). If 4 GB VRAM is available, IQ2_XXS or IQ2_XS (0.19–0.20 GB file) are suitable. For higher quality at the cost of memory, Q3_K_M or Q4_K_M (0.24–0.27 GB file, ~1.1 GB memory) can be used on 12 GB VRAM systems.
What GPU do you need for Kimi K3 0.40B?
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 Q6_K (0.3 GB file, about 1.1 GB in use).
Which Kimi K3 0.40B quant should you choose?
Q4_K_M is the usual starting point for Kimi K3 0.40B: it balances file size and quality well. Here it is a 0.3 GB file.
If memory is tight, the smallest file is the first one in the table at 0.0 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 Kimi K3 0.40B
How much VRAM does Kimi K3 0.40B need?
About 1.1 GB at Q4_K_M with an 8K context window. That is the 0.3 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 Kimi K3 0.40B on an 8 GB GPU?
Yes, with Q6_K: the 0.3 GB file needs about 1.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Kimi K3 0.40B GGUF?
It depends on the quant: file is 0.0 GB, IQ1_S is 0.2 GB, IQ1_M is 0.2 GB, IQ2_XXS is 0.2 GB, IQ2_XS is 0.2 GB, IQ2_S is 0.2 GB and 19 more.
What context length does Kimi K3 0.40B support?
The GGUF header lists a trained context length of 4,096 tokens (4K). The KV cache grows with the context you use: about 0.3 GB at 8K and 0.2 GB at 4K.
Where can I download Kimi K3 0.40B GGUF?
The files are on Hugging Face in the mradermacher/Kimi-K3-0.40B-i1-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 Kimi K3 0.40B 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 Kimi K3 0.40B on 8 GB VRAM?
Yes, using IQ1_S or IQ1_M quantizations requires about 1 GB VRAM at 8 k context.
Does this model support vision tasks?
The vision tower is present but untrained; the model is not suitable for vision inference.
What license applies to this model?
The model is released under the MIT license.