Inference · compare

GLM-4.7 Flash vs gpt-oss 20B.

Side by side only on machines of the same kind, so the difference is the model, not the computer. Each figure is the fastest build of that model on that machine.

1machine kind both ran on
MachineGLM-4.7 Flash tok/sgpt-oss 20B tok/sContext A / B
2× Tesla T4 · 15 GB45Q4_K_M (4-bit) by unsloth55Q3_K_S (3-bit) by unsloth128K / 64K