How fast is gpt-oss-120b on Strix Halo?

The fastest of 6 measured configs is 53.7 tok/s single-stream (MXFP4, llama.cpp/vulkan). The slowest is 17.8 tok/s.

Which backend is fastest for gpt-oss-120b on Strix Halo?

llama.cpp/vulkan at 53.7 tok/s, then llama.cpp/rocm at 50.2. Each backend figure is its best plain-decoding row and may use a different quant.

Configs › gpt-oss-120b › gpt-oss-120b on Strix Halo: 53.7 tok/s, 6 measured configs

gpt-oss-120b on Strix Halo: 53.7 tok/s, 6 measured configs

Measured gpt-oss-120b speeds on Strix Halo: fastest 53.7 tok/s single-stream (MXFP4, llama.cpp/vulkan), 6 configs with quant, backend and context, each linked to its source, snapshot 2026-10-29.

Snapshot 2026-10-29 · 413 configs on AI boxes and 288 on GPU cards · 121 models. Every number is generated from the tracker snapshot and linked to where it was measured; nothing here is typed by hand.

Fastest measured single-stream run of gpt-oss-120b on Strix Halo: 53.7 tok/s (MXFP4, llama.cpp/vulkan), snapshot 2026-10-29.

Every measured single-stream run of gpt-oss-120b on Strix Halo

Decode tok/s	Quant	Backend	Context	Trust	Measured by	Source checked	Source
53.7	MXFP4	llama.cpp/vulkan		Extracted, source-linked	huppiflupp	2026-10-24	source
52.9	MXFP4	llama.cpp/vulkan	128	Extracted, source-linked	wireframeslayout	2026-09-06	source
50.6	MXFP4	llama.cpp/vulkan	512	Structured table	hogeheer	2026-10-29	source
50.2	MXFP4	llama.cpp/rocm	128	Extracted, source-linked	wireframeslayout	2026-09-06	source
31.5	MXFP4	llama.cpp/vulkan	32768	Extracted, source-linked	lhl	2026-10-14	source
17.8	MXFP4	llama.cpp/rocm	32768	Extracted, source-linked	lhl	2026-10-11	source

Backends on Strix Halo

Fastest plain-decoding run per backend. Quants can differ between rows.

llama.cpp/vulkan: 53.7 tok/s (MXFP4)
llama.cpp/rocm: 50.2 tok/s (MXFP4)

Questions and answers

How fast is gpt-oss-120b on Strix Halo?: The fastest of 6 measured configs is 53.7 tok/s single-stream (MXFP4, llama.cpp/vulkan). The slowest is 17.8 tok/s.
Which backend is fastest for gpt-oss-120b on Strix Halo?: llama.cpp/vulkan at 53.7 tok/s, then llama.cpp/rocm at 50.2. Each backend figure is its best plain-decoding row and may use a different quant.

See also: gpt-oss-120b on every box · Strix Halo hardware page (from S$4,329 in Singapore today).

Last updated: snapshot 2026-10-29; newest source check 2026-10-29.

More: all configs · hardware · news · contributors · methodology · llms.txt · llms-full.txt · API (OpenAPI) · JSON snapshot. Built by Altronis, private on-prem AI, Singapore.