KI

Kimi Linear 48B A3B Instruct

KimiGGUF by bartowskiUpdated 2026-02-09
Download on Hugging Face
Parameters49.1B
Context1024K
Layers27
Architecturekimi-linear
Smallest quant0.2 GB
Quick answer

Kimi Linear 48B A3B Instruct needs about 39.7 GB of memory at Q4_K_M (a 30.1 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 30 quants, from 0.2 GB to 98.3 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 36.7 GB.

Kimi Linear 48B A3B Instruct – large‑scale linear attention model

Kimi Linear 48B A3B Instruct is a 48 billion‑parameter model with a 1 048 576‑token context. It uses a hybrid KDA‑to‑MLA architecture and is trained on 5.7 T tokens. The model is released under the MIT license and is available in multiple GGUF quantizations.

Good at

  • Long‑context inference up to 1 M tokens
  • Efficient decoding with reduced KV cache
  • High throughput on modern GPUs

Limits

  • Requires 8 k context for memory estimates
  • Large VRAM needed for higher‑quality quants
  • No vision or multimodal support

Best for

  • Chat and instruction following at long context
  • Fine‑tuning on specialized corpora
  • Research on linear attention mechanisms

Who it’s for. Engineers and researchers who need a large‑scale model with efficient long‑context performance. The model is open‑source and can be run with llama.cpp, vllm, or other compatible tools.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
—0.2 GB 9.8 GB
Too big Get file →
IQ1_S10.5 GB 20.2 GB
Too big Get file →
IQ1_M10.9 GB 20.6 GB
Too big Get file →
IQ2_XXS12.1 GB 21.8 GB
Too big Get file →
IQ2_XS13.9 GB 23.6 GB
Too big Get file →
IQ2_S14.0 GB 23.6 GB
Too big Get file →
IQ2_M15.8 GB 25.5 GB
Too big Get file →
Q2_K17.5 GB 27.2 GB
Too big Get file →
Q2_K_L17.9 GB 27.6 GB
Too big Get file →
IQ3_XXS19.7 GB 29.3 GB
Too big Get file →
IQ3_XS20.4 GB 30.1 GB
Too big Get file →
Q3_K_S21.6 GB 31.3 GB
Too big Get file →
IQ3_M22.7 GB 32.3 GB
Too big Get file →
Q3_K_M22.7 GB 32.4 GB
Too big Get file →
Q3_K_L23.6 GB 33.3 GB
Too big Get file →
Q3_K_XL23.9 GB 33.6 GB
Too big Get file →
IQ4_XS26.5 GB 36.1 GB
Too big Get file →
IQ4_NL28.0 GB 37.6 GB
Too big Get file →
Q4_028.4 GB 38.1 GB
Too big Get file →
Q4_K_S29.0 GB 38.7 GB
Too big Get file →
Q4_K_M30.1 GB 39.7 GB
Too big Get file →
Q4_K_L30.3 GB 40.0 GB
Too big Get file →
Q4_131.0 GB 40.6 GB
Too big Get file →
Q5_K_S34.0 GB 43.7 GB
Too big Get file →
Q5_K_M35.1 GB 44.8 GB
Too big Get file →
Q5_K_L35.3 GB 45.0 GB
Too big Get file →
Q6_K40.5 GB 50.1 GB
Too big Get file →
Q6_K_L40.6 GB 50.3 GB
Too big Get file →
Q8_052.2 GB (2 parts) 61.9 GB
Too big Get file →
BF1698.3 GB (3 parts) 108.0 GB
Too big Get file →

For 8 GB VRAM, use IQ1_S (10.49 GB file, ~20 GB VRAM at 8 k context). For 16 GB VRAM, IQ2_M (15.79 GB file, ~25.5 GB VRAM). For 24 GB VRAM, Q4_K_M (30.06 GB file, ~39.7 GB VRAM). Estimates are approximate.

What GPU do you need for Kimi Linear 48B A3B Instruct?

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: the best quant that fits is the main file (0.2 GB file, about 9.8 GB in use).
  • 16 GB VRAM: the best quant that fits is the main file (0.2 GB file, about 9.8 GB in use).
  • 24 GB VRAM: the best quant that fits is IQ2_S (14.0 GB file, about 23.6 GB in use).
  • 32 GB VRAM: the best quant that fits is Q3_K_S (21.6 GB file, about 31.3 GB in use).
  • 48 GB VRAM: the best quant that fits is Q5_K_L (35.3 GB file, about 45.0 GB in use).

Which Kimi Linear 48B A3B Instruct quant should you choose?

Q4_K_M is the usual starting point for Kimi Linear 48B A3B Instruct: it balances file size and quality well. Here it is a 30.1 GB file.

The Q8_0 file is 52.2 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

Architecturekimi-linear
Parameters49.1B (approximate, mixture of experts)
Active parameters3B
Layers27
Trained context1,048,576 tokens
KV cache (fp16)1094 KB per token

Source

Quantized bybartowski
FamilyKimi
Licensemit
Downloads (HF)18,758
Last updated2026-02-09

Frequently asked questions about Kimi Linear 48B A3B Instruct

How much VRAM does Kimi Linear 48B A3B Instruct need?

About 39.7 GB at Q4_K_M with an 8K context window. That is the 30.1 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 Linear 48B A3B Instruct on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 9.8 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Kimi Linear 48B A3B Instruct GGUF?

It depends on the quant: file is 0.2 GB, IQ1_S is 10.5 GB, IQ1_M is 10.9 GB, IQ2_XXS is 12.1 GB, IQ2_XS is 13.9 GB, IQ2_S is 14.0 GB and 24 more.

What context length does Kimi Linear 48B A3B Instruct support?

The GGUF header lists a trained context length of 1,048,576 tokens (1024K). The KV cache grows with the context you use: about 9.2 GB at 8K and 36.7 GB at 32K.

Where can I download Kimi Linear 48B A3B Instruct GGUF?

The files are on Hugging Face in the bartowski/moonshotai_Kimi-Linear-48B-A3B-Instruct-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 Linear 48B A3B Instruct 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 Linear 48B on 8 GB VRAM?

Use the IQ1_S quant; it requires about 20 GB VRAM at 8 k context, so 8 GB VRAM is insufficient.

Does the model support vision or multimodal input?

No, the model only handles text; it has no vision or multimodal capabilities.

What is the maximum context length I can use?

The model was trained with a context of 1 048 576 tokens, but practical limits depend on VRAM and the chosen quant.

Other GGUFs of Kimi-Linear-48B-A3B-Instruct

Similar models

Browse all →

More from bartowski

All by bartowski →

More Kimi models

All Kimi →

Other large models

All large models →