RunsOnMy ModelsGPUs & MacsBy VRAMBest GPUCan I run…?OllamaStart hereAPI pricesVRAM calculatorNew

Can Kimi VL A3B Thinking 2506 run on the RTX 3080 10GB?

Yes. At 8K context the best fit is Q3_K_M, using about 8.4 GB of the 9.5 GB usable on the RTX 3080 10GB. Estimated speed ceiling: ~518 tokens/sec.

Better quality with offload: 4-bit (Q4_K_M)

Runs with offload 4-bit (Q4_K_M) at 8K context: all attention layers plus the experts of 20 of 26 MoE layers on the RTX 3080 10GB (about 8.1 GB), and the experts of the other 6 layers in system RAM (about 2.0 GB). Needs a PC with at least 32 GB of system RAM. Estimated speed: 81–127 tokens/sec (usable for chat), versus a bandwidth ceiling of ~394 tok/s if it fit fully in VRAM.

llama-server -hf ggml-org/Kimi-VL-A3B-Thinking-2506-GGUF:Q4_K_M -c 8192 -ngl 999 --n-cpu-moe 7 --no-mmproj

Estimate, not a benchmark. The command downloads the GGUF from Hugging Face. The speed range spans dual-channel DDR4-3200 to DDR5-6000 system RAM; your CPU, RAM speed and llama.cpp version move it. If llama.cpp runs out of GPU memory, raise --n-cpu-moe by 1-2; if VRAM is left over, lower it for more speed. Recent llama.cpp builds can also pick the split automatically. How we estimate offload.

+ 0.6 GB with image input. The figures on this page are for text only. With image input enabled, the vision projector (mmproj) file loads alongside the model: Q3_K_M at 8K context then needs about 9.0 GB (tight). llama.cpp's -hf downloads and loads it automatically unless you add --no-mmproj. Calculator with image input on.

Memory by quant and context

Quant4K ctx8K ctx32K ctx128K ctxMax tok/s (est., 8K)
2-bit (Q2_K)6.6 GB Fits6.7 GB Fits7.4 GB Fits10.2 GB No676
3-bit (Q3_K_M)8.3 GB Fits8.4 GB Fits9.1 GB Tight12.0 GB No518
4-bit (Q4_K_M) •10.6 GB No10.8 GB No11.5 GB No14.3 GB No—
5-bit (Q5_K_M)11.7 GB No11.8 GB No12.5 GB No15.4 GB No—
6-bit (Q6_K)13.4 GB No13.5 GB No14.2 GB No17.0 GB No—
8-bit (Q8_0) •16.6 GB No16.8 GB No17.5 GB No20.3 GB No—
16-bit (F16/BF16) •30.6 GB No30.7 GB No31.4 GB No34.3 GB No—

• real GGUF file size from ggml-org/Kimi-VL-A3B-Thinking-2506-GGUF. Speed is a bandwidth ceiling (760 GB/s), not a benchmark. Methodology.

What fits instead

"Runs" and "fits" mean the whole model in GPU memory at 8K context (4-bit or better for the model pick). Prices are US launch MSRPs, not today's street prices. "Popular" means the 60 most-downloaded models we track.

More

Kimi VL A3B Thinking 2506 on every GPU and Mac · Every model that runs on the RTX 3080 10GB · Other models that fit 12 GB VRAM · Try other settings in the calculator

Guides: Best GPU for local LLMs by budget · Can I run gpt-oss, Llama 70B or Qwen3? · Mac vs PC for local LLMs

Data: Hugging Face Hub (moonshotai/Kimi-VL-A3B-Thinking-2506), NVIDIA specs. Model data last checked 2026-11-18; refreshed hourly.