The files can be run locally through Unsloth Desktop or a compatible llama.cpp build, and are available from the unsloth/MiniMax-M3.1-Flash-GGUF repository.
The documented variants range from the 129.4 GB UD-IQ1_M file to the 259.1 GB UD-Q8_K_XL file. For a 16K text context and one parallel slot, Unsloth estimates combined RAM and VRAM requirements of 134 GB for 1-bit, 150 GB for 2-bit, 176–208 GB for 3-bit, 256 GB for 4-bit, and 272 GB for 8-bit. Its example configuration uses the 143.9 GB UD-Q2_K_XL quant, high reasoning effort, and CPU/GPU offloading as needed.
Unsloth reports its own top-1 agreement and KLD comparisons across seven categories; for example, it lists 88.5% general-chat agreement for Q2 and 97.3% for Q4. The documentation does not establish independent benchmark performance, real-world throughput, or that these requirements will fit every hardware and runtime configuration.