RunsOnMy ModelsGPUs & MacsBy VRAMAPI pricesVRAM calculatorNew

Can I run Kimi VL A3B Thinking 2506 locally?

Kimi VL A3B Thinking 2506 by moonshotai has 16B parameters (3.0B active per token, 64 experts, 6 used per token). At 4-bit with an 8K context it needs about 10.8 GB of GPU memory. It fits on 65 of 65 GPUs and Macs we track. See all models that fit in 12 GB of VRAM.

10.8 GBQ4_K_M · 8K context
RTX 4060 Ti 16GBsmallest consumer GPU that fits
M2 24GBsmallest Mac that fits
128Kmax context
MLAattention type

Memory needed by quant and context length

GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use real file sizes from ggml-org/Kimi-VL-A3B-Thinking-2506-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.

QuantQuality4K ctx8K ctx32K ctx128K ctx
2-bit (Q2_K)rough6.6 GB6.7 GB7.4 GB10.2 GB
3-bit (Q3_K_M)usable8.3 GB8.4 GB9.1 GB12.0 GB
4-bit (Q4_K_M) •good, most popular10.6 GB10.8 GB11.5 GB14.3 GB
5-bit (Q5_K_M)very good11.7 GB11.8 GB12.5 GB15.4 GB
6-bit (Q6_K)near lossless13.4 GB13.5 GB14.2 GB17.0 GB
8-bit (Q8_0) •effectively lossless16.6 GB16.8 GB17.5 GB20.3 GB
16-bit (F16/BF16) •original weights60.3 GB60.4 GB61.1 GB64.0 GB

Which GPUs and Macs run Kimi VL A3B Thinking 2506?

Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).

HardwareMemoryFits?Best quantNeedsMax tok/s (est.)
RTX 3060 12GB12 GBFitsQ3_K_M8.4 GB246details
RTX 4060 Ti 16GB16 GBFitsQ6_K13.5 GB117details
RTX 309024 GBFitsQ8_016.8 GB302details
RTX 409024 GBFitsQ8_016.8 GB325details
RTX 509032 GBFitsQ8_016.8 GB577details
RTX 5070 Ti16 GBFitsQ6_K13.5 GB364details
RX 7900 XTX24 GBFitsQ8_016.8 GB309details
Mac M4 Pro 48GB48 GBFitsQ8_016.8 GB88details
Mac M4 Max 128GB128 GBFitsF1660.4 GB47details
Mac M3 Ultra 512GB512 GBFitsF1660.4 GB70details
DGX Spark128 GBFitsF1660.4 GB23details
RTX 407012 GBFitsQ3_K_M8.4 GB344details
RTX 5060 Ti 16GB16 GBFitsQ6_K13.5 GB182details
RTX 40608 GBFitsQ2_K6.7 GB242details
Arc B58012 GBFitsQ3_K_M8.4 GB311details
RTX 3060 Ti8 GBFitsQ2_K6.7 GB398details
RTX 30708 GBFitsQ2_K6.7 GB398details
RTX 3080 10GB10 GBFitsQ3_K_M8.4 GB518details
RTX 3080 12GB12 GBFitsQ3_K_M8.4 GB622details
RTX 3090 Ti24 GBFitsQ8_016.8 GB325details
RTX 4060 Ti 8GB8 GBFitsQ2_K6.7 GB256details
RTX 4070 Super12 GBFitsQ3_K_M8.4 GB344details
RTX 4070 Ti12 GBFitsQ3_K_M8.4 GB344details
RTX 4070 Ti Super16 GBFitsQ6_K13.5 GB273details
RTX 408016 GBFitsQ6_K13.5 GB291details
RTX 4080 Super16 GBFitsQ6_K13.5 GB299details
RTX 50608 GBFitsQ2_K6.7 GB398details
RTX 507012 GBFitsQ3_K_M8.4 GB458details
RTX 508016 GBFitsQ6_K13.5 GB390details
2x RTX 30902×24 GBFitsQ8_016.8 GB302details
2x RTX 40902×24 GBFitsQ8_016.8 GB325details
2x RTX 50902×32 GBFitsQ8_016.8 GB577details
RTX A600048 GBFitsQ8_016.8 GB247details
RTX 6000 Ada48 GBFitsQ8_016.8 GB309details
RTX PRO 6000 Blackwell96 GBFitsF1660.4 GB153details
L40S48 GBFitsQ8_016.8 GB278details
A100 40GB40 GBFitsQ8_016.8 GB501details
A100 80GB80 GBFitsF1660.4 GB175details
H100 PCIe80 GBFitsF1660.4 GB171details
H100 SXM80 GBFitsF1660.4 GB287details
H200141 GBFitsF1660.4 GB411details
RX 76008 GBFitsQ2_K6.7 GB256details
RX 7700 XT12 GBFitsQ3_K_M8.4 GB295details
RX 7800 XT16 GBFitsQ6_K13.5 GB254details
RX 7900 GRE16 GBFitsQ6_K13.5 GB234details
RX 7900 XT20 GBFitsQ8_016.8 GB258details
RX 907016 GBFitsQ6_K13.5 GB260details
RX 9070 XT16 GBFitsQ6_K13.5 GB260details
Ryzen AI Max+ 395 128GB128 GBFitsF1660.4 GB22details
Arc A770 16GB16 GBFitsQ6_K13.5 GB228details
Mac M1 16GB16 GBFitsQ3_K_M8.4 GB46details
Mac M2 24GB24 GBFitsQ6_K13.5 GB41details
Mac M3 24GB24 GBFitsQ6_K13.5 GB41details
Mac M4 16GB16 GBFitsQ3_K_M8.4 GB82details
Mac M4 32GB32 GBFitsQ8_016.8 GB39details
Mac M1 Pro 32GB32 GBFitsQ8_016.8 GB64details
Mac M3 Pro 36GB36 GBFitsQ8_016.8 GB48details
Mac M4 Pro 24GB24 GBFitsQ6_K13.5 GB111details
Mac M4 Pro 64GB64 GBFitsQ8_016.8 GB88details
Mac M1 Max 64GB64 GBFitsQ8_016.8 GB129details
Mac M2 Max 96GB96 GBFitsF1660.4 GB34details
Mac M3 Max 128GB128 GBFitsF1660.4 GB34details
Mac M4 Max 64GB64 GBFitsQ8_016.8 GB176details
Mac M2 Ultra 192GB192 GBFitsF1660.4 GB69details
Mac M3 Ultra 256GB256 GBFitsF1660.4 GB70details

Questions

How much VRAM does Kimi VL A3B Thinking 2506 need?

About 10.8 GB at 4-bit (Q4_K_M) with an 8K context: 9.8 GB of weights (real GGUF file size), 0.2 GB of KV cache and 0.7 GB of runtime overhead. At 8-bit it needs about 16.8 GB.

What is the smallest GPU that runs Kimi VL A3B Thinking 2506?

At Q4_K_M and 8K context, the smallest single consumer GPU in our table that fits it is the RTX 4060 Ti 16GB (16 GB).

Can I run Kimi VL A3B Thinking 2506 on a Mac?

Yes. At Q4_K_M with 8K context it fits on a Mac M2 24GB or larger, using the default macOS GPU memory limit.

Similar models

Same family or a similar parameter count, with the VRAM each needs at Q4_K_M and 8K context.

Kimi VL A3B Instruct

16B · 10.8 GB at 4-bit

Kimi VL A3B Thinking

16B · 10.3 GB at 4-bit

Moonlight 16B A3B Instruct

16B · 10.8 GB at 4-bit

deepseek moe 16b base

16B · 12.5 GB at 4-bit

LLaDA2.1 mini

16B · 10.3 GB at 4-bit

DeepSeek Coder V2 Lite Instruct

16B · 10.5 GB at 4-bit

Model facts

Hugging Face repomoonshotai/Kimi-VL-A3B-Thinking-2506
Parameters16,407,657,776 (16B) · ~3.0B active (from the model name)
Layers / hidden size27 / 2048
AttentionMLA; 16 heads, 16 KV heads, head dim 128
KV cache per token (FP16)30.4 KiB
Max context131,072 tokens
Licensemit
Released on HFJun 21, 2025
Downloads (30d) / likes12,430 / 389

Sources: Hugging Face model API · config.json (moonshotai/Kimi-VL-A3B-Thinking-2506) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.