The repository is labelled as a 26B-parameter qwen3vl architecture and includes GGUF variants.
The model page lists a Q2_K variant at 8.49 GB and a Q4_K_M variant at 14.6 GB. It says WanGP supports Wan-derived models, Hunyuan Video, Flux, Qwen, Z-Image, and LTX models; its stated features include automatic model downloads matched to a user's architecture. The page also says some WanGP-supported models can run with as little as 6 GB of VRAM, and that older RTX 10XX and 20xx GPUs are supported.
An October 15, 2026 update lists a rank-128 adapter for the TaoMate H3 FL2VA 3-Step profile and LightVAE Lite decoder files in BF16 and INT8 ConvRot. The source does not provide benchmark results, measured H3 memory use, generation quality comparisons, or a complete account of licence terms for every included component.