Which local LLMs run in 128 GB of VRAM?

On this site, 1,036 models have a quantization that fits in 128 GB at 8,192 tokens (estimated). Popular ones include Qwen3.8-27B, Qwen3.8-Flash-Next-GSQ-RCO, Ornith-1.5-9B, Huihui-Qwen3.6-35B-A3B-abliterated-MTP, Ornith-1.5-35B-A3B.

What is the largest model that runs in 128 GB of VRAM?

Among the models listed here, MiniMax-M3.1-Flash-Maya (about 426.2B). With IQ2_XXS it needs an estimated 125 GB at 8,192 tokens. Smaller quantizations lower quality.

Is it the same for a Mac with 128 GB of unified memory?

macOS limits how much memory the GPU can use, so not all 128 GB is available to the model. On a Mac, leave headroom over the required memory.

What happens with a longer context?

The KV cache grows, so the required memory grows too. Each model page shows estimated memory for each context length.

Find by VRAM / unified memory

Local LLMs that run in 128 GB of VRAM

Of the 1,276 models listed on this site, 1,036 have a quantization whose estimated memory fits in 128 GB at an 8,192-token context. For each model we show the largest (most precise) quantization that fits.

Data as of Nov 11, 2026

Models that fit: 1,036
Largest model that fits: MiniMax-M3.1-Flash-Maya 426.2B
128 GB devices (listed here): 8

Popular models that run in 128 GB of VRAM

Sorted by Hugging Face downloads (last 30 days, summed across sources). Not a performance ranking.

1 Qwen3.8-27B 27.3B BF16 56.4 GB
2 Qwen3.8-Flash-Next-GSQ-RCO 176.9B IQ3_S 85.0 GB
3 Ornith-1.5-9B 9.2B BF16 19.8 GB
4 Huihui-Qwen3.6-35B-A3B-abliterated-MTP 35.5B F16 72.3 GB
5 Ornith-1.5-35B-A3B 35.5B BF16 72.4 GB
6 Qwen3.8-Flash-Next-GSQ-RCO-abliterated 176.9B IQ3_S 85.3 GB
7 Huihui-Qwen3.8-27B-abliterated 27.3B BF16 56.4 GB
8 OTel-2.0-Decision-26B-A4B-4bit 25.2B Q8_0 28.4 GB
9 Qwen3.8-27B-iMatrix-NVFP4-MTP NVFP4 18.9 GB
10 Tiel-Coder-35B-A3B-GGUF-MTP UD-Q8_K_XL 40.1 GB

Estimated Memory is estimated at 8,192 tokens with an f16 KV cache.

Third-party evaluations of models that run in 128 GB

Top values from other organizations for the original models before quantization (not this site's rating).

Human preference (LMArena)

Source: LMArena (CC BY 4.0).

Capabilities (Epoch AI)

Source: Epoch Capabilities Index (Epoch AI) (CC BY 4.0).

Devices with 128 GB of VRAM or unified memory

Choose a device to see the models that run on it and estimated generation speed.

All models that fit in 128 GB of VRAM

One entry per model, up to 30, sorted by popularity. For each model we show the largest (most precise) quantization that fits in this memory with a context of at least 8K tokens. The context length is the longest one that quantization fits (for models whose maximum is under 8K, such as embedding models, that maximum). Models that are not for chat or text generation are labelled with their use. "Left" is the memory minus the estimate; under 5% of the memory is marked "tight" (not a guarantee that it fits).

Qwen3.8-Flash-Next-GSQ-RCO IQ3_S Image input

Context length 262,144 Estimated 92.0 GB 45.4 GB left
Qwen3.8-27B BF16

Context length 262,144 Estimated 73.0 GB 64.4 GB left
Ornith-1.5-9B BF16

Context length 262,144 Estimated 28.1 GB 109 GB left
Huihui-Qwen3.6-35B-A3B-abliterated-MTP F16 Image input

Context length 262,144 Estimated 77.5 GB 59.9 GB left
Qwen3.8-Flash-Next-GSQ-RCO-abliterated IQ3_S Image input

Context length 262,144 Estimated 92.3 GB 45.2 GB left
Ornith-1.5-35B-A3B BF16

Context length 262,144 Estimated 77.6 GB 59.9 GB left
Huihui-Qwen3.8-27B-abliterated BF16 Image input

Context length 262,144 Estimated 73.0 GB 64.4 GB left
OTel-2.0-Decision-26B-A4B-4bit Q8_0

Context length 262,144 Estimated 33.6 GB 104 GB left
Qwen3.8-27B-iMatrix-NVFP4-MTP NVFP4 Image input

Context length 262,144 Estimated 35.5 GB 102 GB left
Tiel-Coder-35B-A3B-GGUF-MTP UD-Q8_K_XL Image input

Context length 262,144 Estimated 45.3 GB 92.1 GB left
gemma-4-12B-it BF16 Image input

Context length 262,144 Estimated 29.7 GB 108 GB left
Qwen3.8-27B-Uncensored Q8_0

Context length 262,144 Estimated 47.4 GB 90.0 GB left
Ornith-1.0-9B BF16

Context length 262,144 Estimated 27.6 GB 110 GB left
mxbai-embed-large-v1 F32 Embedding (not for chat)

Context length 512 Estimated 2.41 GB 135 GB left
Qwen3.6-35B-A3B-MTP BF16 Image input

Context length 262,144 Estimated 77.5 GB 59.9 GB left
Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP BF16 Image input

Context length 262,144 Estimated 28.1 GB 109 GB left
MOSS-Music-8B-Instruct F16 Speech / audio

Context length 40,960 Estimated 23.3 GB 114 GB left
Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP Q8_K Image input

Context length 262,144 Estimated 49.8 GB 87.6 GB left
gemma-4-26B-A4B-it BF16 Image input

Context length 262,144 Estimated 57.3 GB 80.2 GB left
Qwen3-VL-8B-Instruct BF16 Image input

Context length 262,144 Estimated 55.9 GB 81.5 GB left
Qwen3.8-Flash-Next Q5_K_M Image input

Context length 32,768 Estimated 137 GB 0.70 GB left (tight)
Qwen3.5-9B BF16 Image input

Context length 262,144 Estimated 28.1 GB 109 GB left
Ornith-1.5-9B-uncensored F16

Context length 262,144 Estimated 27.6 GB 110 GB left
Qwen3.6-35B-A3B BF16 Image input

Context length 262,144 Estimated 77.5 GB 59.9 GB left
Ternary-Bonsai-2-27B F16

Context length 262,144 Estimated 72.2 GB 65.2 GB left
Swift-1.5-Qwen3.8-27B-GSQ-RCO IQ3_S

Context length 262,144 Estimated 30.5 GB 107 GB left
Qwen3.5-4B BF16 Image input

Context length 262,144 Estimated 18.4 GB 119 GB left
Ornith-1.0-35B BF16

Context length 262,144 Estimated 75.9 GB 61.6 GB left
Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER-MAX-MTP Q8_0 Image input

Context length 262,144 Estimated 48.6 GB 88.8 GB left
Dirk-Qwen3.8-27B UD-Q8_K_XL Image input

Context length 262,144 Estimated 49.8 GB 87.6 GB left

Frequently asked questions

Which local LLMs run in 128 GB of VRAM?: On this site, 1,036 models have a quantization that fits in 128 GB at 8,192 tokens (estimated). Popular ones include Qwen3.8-27B, Qwen3.8-Flash-Next-GSQ-RCO, Ornith-1.5-9B, Huihui-Qwen3.6-35B-A3B-abliterated-MTP, Ornith-1.5-35B-A3B.
What is the largest model that runs in 128 GB of VRAM?: Among the models listed here, MiniMax-M3.1-Flash-Maya (about 426.2B). With IQ2_XXS it needs an estimated 125 GB at 8,192 tokens. Smaller quantizations lower quality.
Is it the same for a Mac with 128 GB of unified memory?: macOS limits how much memory the GPU can use, so not all 128 GB is available to the model. On a Mac, leave headroom over the required memory.
What happens with a longer context?: The KV cache grows, so the required memory grows too. Each model page shows estimated memory for each context length.