Check my GPU →

MiniMax H3 (33B) on 32 GB of VRAM

Not at full quality. The 16-bit and 8-bit files are too big for 32 GB; the largest that fits is Q5_K_M (23.9 GB). The biggest open video release of 2026 and the most downloaded ComfyUI repo of the year: 33B parameters, 4–15 second clips with native stereo sound. Two variants (first/last-frame and reference-to-video); load one.

Video model · 33.1BAny 32 GB GPUData 2026-10-01
Runs
Yes, with a compressed file.

Take Q5_K_M (23.9 GB). In use it comes to about 29.7 GB, with 2.3 GB left over on 32 GB. Quality: close; small differences in fine detail.

Best fileQ5_K_M
File size23.9 GB
VRAM needed~29.7 GB
System RAM64 GB+
Memory map · 32 GB GPU29.7 GB / 32 GB
016 GB32 GB
Weights Q5_K_M · 23.9 GBWorking memory · 5.0 GBReserve · 0.8 GBFree · 2.3 GB

Also try the INT8 file (34.0 GB) on NVIDIA GPUs. It does not fit entirely, but a current ComfyUI streams the overflow from RAM and native 8-bit files often beat a small GGUF that does fit. Dynamic VRAM →

01What to download

PartFileFolderSize
ModelMiniMax-H3-FL2VA-Q5_K_M.gguf
Q5_K_M · Abiray/MiniMax-H3-GGUF
models/unet23.9 GBDownload →
Text encoderqwen3vl_32b_minimax_h3_int8_convrot.safetensors
Qwen3-VL 32B INT8 (convrot) · Comfy-Org/MiniMax-H3
models/text_encoders27.1 GBDownload →
VAEminimax_h3_video_vae_fp16.safetensors
MiniMax H3 video VAE FP16 · Comfy-Org/MiniMax-H3
models/vae5.2 GBDownload →
Also neededminimax_h3_audio_vae_fp32.safetensors
MiniMax H3 audio VAE FP32 · Comfy-Org/MiniMax-H3
models/vae0.6 GBDownload →
Total download · keep about the same free on disk56.8 GB

Alternatives: Qwen3-VL 32B NVFP4 (AWQ) (15.7 GB); Qwen3-VL 32B BF16 (51.5 GB). FP4 files are listed only as alternatives unless the GPU is an RTX 50 (Blackwell). Sizes from Hugging Face, 2026-10-01.

System RAM: 64 GB or more. These files take about 56.8 GB of RAM while loaded, 62.8 GB with Windows and ComfyUI. What happens with less. calculated

02Every file, on 32 GB

FileSizeNeededOn this cardQualityDownload
16-bit
SAFETENSORS · Comfy-Org
66.3 GB72.1 GB40.1 GB too bigthe original weightsHugging Face →
Q8_0
GGUF · Abiray
36.0 GB41.8 GB9.8 GB too bigpractically identical to the originalHugging Face →
INT8
SAFETENSORS · Comfy-Org
34.0 GB39.8 GB7.8 GB too bigpractically identical to the originalHugging Face →
Q6_K
GGUF · Abiray
28.2 GB34.0 GB2.0 GB too bigvery close to the originalHugging Face →
Q5_K_M ←
GGUF · Abiray
23.9 GB29.7 GBfits · 2.3 GB spareclose; small differences in fine detailHugging Face →
Q4_K_M
GGUF · Abiray
19.9 GB25.7 GBfits · 6.3 GB sparegood; some loss in fine detail and textHugging Face →
Q3_K_M
GGUF · Abiray
15.6 GB21.4 GBfits · 10.6 GB sparenoticeable loss of detailHugging Face →

“Needed” adds 5 GB of working memory for MiniMax H3 and a 0.8 GB reserve. Sizes 2026-10-01.