M5Stack LLM8850 M.2 (AX8850)

An AX8850 accelerator on an M.2 card, for adding an NPU to another host.

Fastest measured run

27.05tok/s

Qwen2.5 0.5B Instruct · W4A16 · StackFlow · source ↗

For the M5Stack LLM8850 M.2 card. All options

Chip
AX8850
Price
Not verified
Runs
7
Models
7
Sources
1

Where to buy.

Every run on this board.

ModelSizeQuantRuntimeGen tok/sPrompt tok/sTTFT msSource
Qwen2.5 0.5B Instruct500MW4A16StackFlow27.05––source ↗
accessed 2026-10-17
MiniCPM4 0.5B500MW8A16StackFlow21.05–212.91source ↗
accessed 2026-10-17
Qwen2.5 1.5B Instruct1.5BW4A16StackFlow15.06––source ↗
accessed 2026-10-17
DeepSeek-R1-Distill-Qwen 1.5B1.5BW4A16StackFlow13.29––source ↗
accessed 2026-10-17
Qwen3 0.6B600MW8A16StackFlow12.88–670.51source ↗
accessed 2026-10-17
Qwen3-VL 2B Instruct VLM2BW8A16StackFlow7.8––source ↗
accessed 2026-10-17
Qwen3 1.7B1.7BW8A16StackFlow7.38–796.38source ↗
accessed 2026-10-17

Can the M5Stack LLM8850 run…

Have this board? Submit your own run

Stay in the loop.

New boards, new runtimes, new numbers. One email a month, no hype.