Benchmark result

gpt-oss:20b on Apple M4 — 25.3 tok/s

Measured with Ollama 0.34.4 on September 28, 2026.

What the model built

The coding scenario asks for a playable game in a single HTML file. This is exactly what the model returned, unedited; the frame only adds a line that tells this page how tall it is.

How this run compares

gpt-oss:20b on other hardware

155.2tok/s71.2NVIDIA GeForce RTX 4070 Ti SUPER
148.1tok/s70.5NVIDIA GeForce RTX 4070 Ti SUPER
132.4tok/s78.8AMD Radeon RX 9070/9070 XT/9070 GRE · MXFP4_MOE
120.7tok/s76.52× NVIDIA L40S · MXFP4
111.1tok/s63.8AMD Radeon RX 7900 XTX

Reproduce this run

toolOllama 0.34.4

modelgpt-oss:20b (MXFP4)

context8,192

samplingtemperature 1

clientv0.4.85+101

Set those in Ollama, then:

llm-benchmark benchmark --model "gpt-oss:20b" --tool "Ollama"

The client prompts for context length and thinking mode, and for the KV cache dtype on Unsloth Studio. Sampling is left at the model default — the values above are what the tool reported using, not overrides the client sent.