Best local AI models for 128GB VRAM

Ranked by what actually fits at 32K context, computed from real file bytes.

A 128GB card gives you about 119.04 GiB to work with after driver overhead. 16 indexed models fit at 32K context — the largest being lingbot-world-v2-14b-causal-fast at 18.5B parameters in Q8_0.

From the file· fit from summed bytesFrom the file· KV per layer
4GB6GB8GB10GB12GB16GB20GB24GB32GB48GB64GB96GB128GB
allvideo generation 16

Fits in 128GB at 32K context

largest quantization that fits, per model
ModelModalityBest quantParams○Total◐Headroom◐
Wan2.2-Animate-14Bvideo generationQ8_017.3B18.27 GiB100.77 GiB
Wan2.1-I2V-14B-480Pvideo generationBF1616.4B31.83 GiB87.21 GiB
Bernini-Rvideo generationQ8_014.3B29.56 GiB89.48 GiB
Wan2.2-Distill-Modelsvideo generationQ8_014.3B15.19 GiB103.85 GiB
Wan2.2-TI2V-5Bvideo generationQ8_05.0B5.87 GiB113.17 GiB
Wan2.1-T2V-14Bvideo generationBF1614.3B27.89 GiB91.15 GiB
Wan-Dancer-14Bvideo generationQ6_K17.2B27.75 GiB91.29 GiB
Wan2.1-I2V-14B-720Pvideo generationBF1616.4B31.83 GiB87.21 GiB
Wan2.1-VACE-14Bvideo generationBF1617.3B33.14 GiB85.90 GiB
Wan2.2-S2V-14Bvideo generationBF1616.3B31.37 GiB87.67 GiB
Wan2.1-FLF2V-14B-720Pvideo generationBF1616.4B31.84 GiB87.20 GiB
JoyAI-Echovideo generationQ6_K12.2B19.07 GiB99.97 GiB
HunyuanVideo-1.5video generationQ8_08.3B9.22 GiB109.82 GiB
Wan2.2-TI2V-5B-Turbovideo generationQ8_05.0B5.87 GiB113.17 GiB
SkyReels-V2-DF-14B-540Pvideo generationBF1614.3B27.46 GiB91.58 GiB
lingbot-world-v2-14b-causal-fastvideo generationQ8_018.5B19.40 GiB99.64 GiB
○Spec sheet◐Predictedwhat these mean

This page models a generic 128GB accelerator, so it answers what fits rather than how fast it runs. For tokens per second you need a specific card — pick one from hardware, where bandwidth is known.