BF16 GGUF models
Unquantized bfloat16. Original quality and the largest size. 9004 models have a BF16 file. Sizes run from 1M to 2780B parameters. The longest trained context is 10240K tokens.
What is BF16?
Unquantized bfloat16. Original quality and the largest size. On average it stores about 16 bits per weight, so a BF16 file is roughly 100% of the size of the same model in 16-bit. The unquantized model. It gives original quality and needs the most memory, so it is mainly for conversion or reference.
9,004 models have a BF16 file on ggufmodels.com, and those files range from 0.0 to 4893.2 GB. The most downloaded are YuE2 audio tokenizer, the v9 head of Mothersuperior, on MERT-v2-FullSong, Qwen3.8-27B and Ornith-1.5-9B. For a longer explanation of every quant type, read GGUF quantization explained.
Frequently asked questions
What is BF16 in GGUF?
Unquantized bfloat16. Original quality and the largest size. It averages about 16 bits per weight.
Should I use BF16?
The unquantized model. It gives original quality and needs the most memory, so it is mainly for conversion or reference.
How big is a BF16 file?
BF16 files on this site range from 0.0 to 4893.2 GB, depending on the model. As a rule of thumb it is about 100% of the 16-bit size.