How fast is Qwen3-Next 80B-A3B on Strix Halo?

The fastest of 3 measured configs is 54.9 tok/s single-stream (UD-Q4_K_XL, llama.cpp/vulkan). The slowest is 45.1 tok/s.

Which backend is fastest for Qwen3-Next 80B-A3B on Strix Halo?

llama.cpp/vulkan at 54.9 tok/s, then llama.cpp/rocm at 49.6, ollama at 45.1. Each backend figure is its best plain-decoding row and may use a different quant.

Configs › Qwen3-Next 80B-A3B › Qwen3-Next 80B-A3B on Strix Halo: 54.9 tok/s, 3 measured configs

Qwen3-Next 80B-A3B on Strix Halo: 54.9 tok/s, 3 measured configs

Measured Qwen3-Next 80B-A3B speeds on Strix Halo: fastest 54.9 tok/s single-stream (UD-Q4_K_XL, llama.cpp/vulkan), 3 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 Qwen3-Next 80B-A3B on Strix Halo: 54.9 tok/s (UD-Q4_K_XL, llama.cpp/vulkan), snapshot 2026-10-29.

Every measured single-stream run of Qwen3-Next 80B-A3B on Strix Halo

Decode tok/s	Quant	Backend	Context	Trust	Measured by	Source checked	Source
54.9	UD-Q4_K_XL	llama.cpp/vulkan	512	Structured table	hogeheer	2026-10-29	source
49.6	UD-Q4_K_XL	llama.cpp/rocm	512	Structured table	hogeheer	2026-10-29	source
45.1		ollama		Extracted, source-linked	huppiflupp	2026-10-24	source

Backends on Strix Halo

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

llama.cpp/vulkan: 54.9 tok/s (UD-Q4_K_XL)
llama.cpp/rocm: 49.6 tok/s (UD-Q4_K_XL)
ollama: 45.1 tok/s

Questions and answers

How fast is Qwen3-Next 80B-A3B on Strix Halo?: The fastest of 3 measured configs is 54.9 tok/s single-stream (UD-Q4_K_XL, llama.cpp/vulkan). The slowest is 45.1 tok/s.
Which backend is fastest for Qwen3-Next 80B-A3B on Strix Halo?: llama.cpp/vulkan at 54.9 tok/s, then llama.cpp/rocm at 49.6, ollama at 45.1. Each backend figure is its best plain-decoding row and may use a different quant.

See also: Qwen3-Next 80B-A3B 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.