gpt-oss 120B

117B total / 5.1B active · load estimate uses total • 71.2GB VRAM (Q4 estimate) • 128,000 context

Estimate · Q4_K_M · 4k context · batch 1. For dense 7B–70B this sits ~5–8% above the Q4_K_M GGUF file. Not peak runtime VRAM and not a measured bench.

Native weights ship as MXFP4. The Q4_K_M figure is the shared formula on total parameters — not the MXFP4 pack size.