Open-weight local model

OpenAI gpt-oss-20b

Ships in native MXFP4. The estimate follows that checkpoint rather than treating it as generic GGUF Q4.

Parameters: 21.512B
Recommended quantization: MXFP4
Weight estimate: 12.8GB
Maximum context: 128K

Official model card

Quantization memory at the default context

Quantization	Weights	KV cache	Total VRAM
MXFP4	12.80GB	0.20GB	14.28GB

GPUs to check for this model