Kimi VL A3B Instruct — Hardware Requirements & GPU Compatibility
VisionFunctions
Kimi VL A3B Instruct is Moonshot AI's 16.4-billion-parameter mixture-of-experts vision-language model, with about 3 billion parameters active per token (the A3B in its name). Because only the active experts run for each token, it runs faster than a dense model of similar total size, though the full set of weights must still fit in memory. It pairs a native-resolution vision encoder with a language core built on Moonshot's Moonlight-16B-A3B model, supporting image and video understanding, OCR, multi-image reasoning, and agent-style tool use. At this size, local inference is practical on a single high-end consumer GPU once quantized. The model supports a 131K token context window, useful for long documents or video transcripts. It is released under the MIT license, one of the most permissive open licenses, and was published in April 2025.
200.0K downloads
284 likesSpecifications
- Publisher
- Moonshot AI
- Family
- Kimi
- Parameters
- 16.4B
- Architecture
- KimiVLForConditionalGeneration
- Context Length
- 131,072 tokens
- Vocabulary Size
- 163,840
- Release Date
- 2025-04-09
- License
- MIT
Get Started
HuggingFace
How Much VRAM Does Kimi VL A3B Instruct Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| Q2_K | 3.40 | 7.7 GB | 36.3 GB | 6.97 GB | 2-bit quantization with K-quant improvements |
| Q3_K_S | 3.50 | 7.9 GB | 36.5 GB | 7.18 GB | 3-bit small quantization |
| Q3_K_M | 3.90 | 8.8 GB | 37.3 GB | 8.00 GB | 3-bit medium quantization |
| Q4_0 | 4.00 | 9.0 GB | 37.5 GB | 8.20 GB | 4-bit legacy quantization |
| Q4_K_M | 4.80 | 10.6 GB | 39.1 GB | 9.84 GB | 4-bit medium quantization — most popular sweet spot |
| Q5_K_M | 5.70 | 12.4 GB | 41.0 GB | 11.69 GB | 5-bit medium quantization — good quality/size tradeoff |
| Q6_K | 6.60 | 14.3 GB | 42.8 GB | 13.54 GB | 6-bit quantization, very good quality |
| Q8_0 | 8.00 | 17.2 GB | 45.7 GB | 16.41 GB | 8-bit quantization, near-lossless |
Which GPUs Can Run Kimi VL A3B Instruct?
Q4_K_M · 10.6 GBKimi VL A3B Instruct (Q4_K_M) requires 10.6 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 14+ GB is recommended. Using the full 131K context window can add up to 28.5 GB, bringing total usage to 39.1 GB. 38 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3080 Ti.
Runs great
— Plenty of headroom
Decent
— Enough VRAM, may be tightWhich Devices Can Run Kimi VL A3B Instruct?
Q4_K_M · 10.6 GB48 devices with unified memory can run Kimi VL A3B Instruct, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, NVIDIA Jetson Orin NX 16GB.
Runs great
— Plenty of headroomDecent
— Enough memory, may be tightWhere to Download Kimi VL A3B Instruct
Community quantizations of this model — GGUF for llama.cpp, Ollama, and LM Studio, plus AWQ/MLX variants where available.
Related Models
Frequently Asked Questions
- How much VRAM does Kimi VL A3B Instruct need?
Kimi VL A3B Instruct requires 10.6 GB of VRAM at Q4_K_M, or 33.6 GB at BF16. Full 131K context adds up to 28.5 GB (39.1 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 16.4B × 4.8 bits ÷ 8 = 9.8 GB
KV Cache + Overhead ≈ 0.8 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 29.3 GB (at full 131K context)
VRAM usage by quantization
Q4_K_M10.6 GBQ4_K_M + full context39.1 GB- Can NVIDIA GeForce RTX 4090 run Kimi VL A3B Instruct?
Yes, at Q8_0 (17.2 GB) or lower. Higher quantizations like BF16 (33.6 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
- What's the best quantization for Kimi VL A3B Instruct?
For Kimi VL A3B Instruct, Q4_K_M (10.6 GB) offers the best balance of quality and VRAM usage. Q5_K_S (12.0 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 5.3 GB.
VRAM requirement by quantization
IQ2_XXS5.3 GBIQ3_XS7.5 GBQ4_09.0 GBIQ4_NL10.0 GBQ4_K_M ★10.6 GBBF1633.6 GB★ Recommended — best balance of quality and VRAM usage.
- Can I run Kimi VL A3B Instruct on a Mac?
Kimi VL A3B Instruct requires at least 5.3 GB at IQ2_XXS, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
- Can I run Kimi VL A3B Instruct locally?
Yes — Kimi VL A3B Instruct can run locally on consumer hardware. At Q4_K_M quantization it needs 10.6 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is Kimi VL A3B Instruct?
At Q4_K_M, Kimi VL A3B Instruct can reach ~171 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~208 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 10.6 × 0.65 = ~526 tok/s
Estimated speed at Q4_K_M (10.6 GB)
~526 tok/s~208 tok/s~526 tok/s~462 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of Kimi VL A3B Instruct?
At Q4_K_M, the download is about 9.84 GB. The full-precision BF16 version is 32.82 GB. The smallest option (IQ2_XXS) is 4.51 GB.
- Which GPUs can run Kimi VL A3B Instruct?
38 consumer GPUs can run Kimi VL A3B Instruct at Q4_K_M (10.6 GB). Top options include AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 6900 XT, AMD Radeon RX 6700 XT. 26 GPUs have plenty of headroom for comfortable inference.
- Which devices can run Kimi VL A3B Instruct?
52 devices with unified memory can run Kimi VL A3B Instruct at Q4_K_M (10.6 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Apple iPhone 17 Pro, Asus ROG Flow Z13 (2025, Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.