moonshotai Kimi K2 Thinking needs about 625.6 GB of memory at Q4_K_M (a 624.0 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 24 quants, from 1.5 GB to 1091.1 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.
Kimi K2 Thinking – a 1.03‑billion‑parameter deepseek2 model focused on step‑by‑step reasoning and tool orchestration
Kimi K2 Thinking is a 1.03‑billion‑parameter model built on the deepseek2 architecture. It is trained for chain‑of‑thought reasoning and dynamic tool invocation, supporting up to 262,144 context tokens. The model is available in multiple GGUF quantizations ranging from IQ1_S to Q8_0.
Good at
- Step‑by‑step reasoning
- Tool orchestration
- Long‑horizon agentic tasks
Limits
- Requires large VRAM for high‑precision quants
- No vision support
- License is other, not open source
Best for
- Research workflows needing multi‑step reasoning
- Coding assistants with tool calls
- Long‑form content generation
Who it’s for. Developers and researchers who need a reasoning‑centric LLM that can invoke external tools over many steps. The model is suitable for building autonomous agents and complex workflows.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 1.5 GB | 3.1 GB | Fits | Get file → | |
| IQ1_S | 210.6 GB (6 parts) | 212.2 GB | Too big | Get file → | |
| IQ1_M | 219.9 GB (6 parts) | 221.5 GB | Too big | Get file → | |
| IQ2_XXS | 245.6 GB (7 parts) | 247.2 GB | Too big | Get file → | |
| IQ2_XS | 283.4 GB (8 parts) | 285.0 GB | Too big | Get file → | |
| IQ2_S | 284.1 GB (8 parts) | 285.7 GB | Too big | Get file → | |
| IQ2_M | 323.3 GB (9 parts) | 324.9 GB | Too big | Get file → | |
| Q2_K | 359.0 GB (10 parts) | 360.6 GB | Too big | Get file → | |
| Q2_K_L | 360.1 GB (10 parts) | 361.7 GB | Too big | Get file → | |
| IQ3_XXS | 405.2 GB (11 parts) | 406.8 GB | Too big | Get file → | |
| IQ3_XS | 420.7 GB (11 parts) | 422.3 GB | Too big | Get file → | |
| Q3_K_S | 445.0 GB (12 parts) | 446.6 GB | Too big | Get file → | |
| IQ3_M | 467.5 GB (13 parts) | 469.1 GB | Too big | Get file → | |
| Q3_K_M | 467.5 GB (13 parts) | 469.1 GB | Too big | Get file → | |
| Q3_K_L | 487.7 GB (13 parts) | 489.3 GB | Too big | Get file → | |
| Q3_K_XL | 488.7 GB (13 parts) | 490.3 GB | Too big | Get file → | |
| IQ4_XS | 548.2 GB (14 parts) | 549.8 GB | Too big | Get file → | |
| IQ4_NL | 580.0 GB (16 parts) | 581.6 GB | Too big | Get file → | |
| Q4_0 | 590.1 GB (16 parts) | 591.7 GB | Too big | Get file → | |
| Q4_K_M | 624.0 GB (16 parts) | 625.6 GB | Too big | Get file → | |
| Q4_1 | 643.3 GB (17 parts) | 644.9 GB | Too big | Get file → | |
| Q5_K_M | 730.5 GB (20 parts) | 732.1 GB | Too big | Get file → | |
| Q6_K | 844.3 GB (23 parts) | 845.9 GB | Too big | Get file → | |
| Q8_0 | 1091.1 GB (30 parts) | 1092.6 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_0 (file size 590.11 GB, memory 591.7 GB at 8 k context). For 12 GB VRAM, Q4_K_M (file size 624.02 GB, memory 625.6 GB). For 16 GB VRAM, Q5_K_M (file size 730.53 GB, memory 732.1 GB). These estimates are approximate.
What GPU do you need for moonshotai Kimi K2 Thinking?
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 (1.5 GB file, about 3.1 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (1.5 GB file, about 3.1 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (1.5 GB file, about 3.1 GB in use).
- 12 GB VRAM: the best quant that fits is the main file (1.5 GB file, about 3.1 GB in use).
- 16 GB VRAM: the best quant that fits is the main file (1.5 GB file, about 3.1 GB in use).
- 24 GB VRAM: the best quant that fits is the main file (1.5 GB file, about 3.1 GB in use).
- 32 GB VRAM: the best quant that fits is the main file (1.5 GB file, about 3.1 GB in use).
- 48 GB VRAM: the best quant that fits is the main file (1.5 GB file, about 3.1 GB in use).
Which moonshotai Kimi K2 Thinking quant should you choose?
Q4_K_M is the usual starting point for moonshotai Kimi K2 Thinking: it balances file size and quality well. Here it is a 624.0 GB file.
The Q8_0 file is 1091.1 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 1.5 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 moonshotai Kimi K2 Thinking
How much VRAM does moonshotai Kimi K2 Thinking need?
About 625.6 GB at Q4_K_M with an 8K context window. That is the 624.0 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 moonshotai Kimi K2 Thinking on an 8 GB GPU?
Yes, with the smaller files: the 1.5 GB file needs about 3.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of moonshotai Kimi K2 Thinking GGUF?
It depends on the quant: file is 1.5 GB, IQ1_S is 210.6 GB, IQ1_M is 219.9 GB, IQ2_XXS is 245.6 GB, IQ2_XS is 283.4 GB, IQ2_S is 284.1 GB and 18 more.
What context length does moonshotai Kimi K2 Thinking support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.
Where can I download moonshotai Kimi K2 Thinking GGUF?
The files are on Hugging Face in the bartowski/moonshotai_Kimi-K2-Thinking-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 moonshotai Kimi K2 Thinking use?
Hugging Face lists the license as other. 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 K2 Thinking on 8 GB VRAM?
Yes, the Q4_0 quant uses about 591.7 GB of memory at 8 k context, which fits in 8 GB VRAM with efficient loading.
Does the model support vision or image input?
No, the model has no vision capability and only processes text.
What is the maximum context length I can use?
The model was trained with 262,144 context tokens, but practical limits depend on the quant and hardware.