Boards / M5Stack Module LLM
M5Stack Module LLM (AX630C): LLM benchmarks
A tiny 54mm module with an Axera AX630C NPU, running StackFlow.
At a glance
- Measured runs11
- Models tested6
- Fastest run12.52 tok/s
- Qwen2.5 0.5B Instruct, W4A16, StackFlow · source ↗
- Sources2
Where to buy
- M5Stack LLM Module Kit (AX630C): $79.9 regular price; out of stock when checked, source ↗ (2026-10-17) Check price on Amazon
- M5Stack Module LLM (as tested in paper): $99.9 price quoted in arXiv paper, source ↗ (2026-04-24) Check price on Amazon
Every run on this board
| Model | Quant | Runtime | Gen tok/s | Prompt tok/s | TTFT ms | Source |
|---|---|---|---|---|---|---|
| Qwen2.5 0.5B Instruct | W4A16 | StackFlow | 12.52 | – | 442.95 | source ↗ accessed 2026-10-17 |
| Qwen2.5 0.5B Instruct | W8A16 | StackFlow | 10.32 | – | 359.8 | source ↗ accessed 2026-10-17 |
| Qwen3 0.6B | W8A16 | StackFlow | 10.28 | – | 361.81 | source ↗ accessed 2026-10-17 |
| InternVL3 1B | W8A16 | StackFlow | 9.78 | – | 534.95 | source ↗ accessed 2026-10-17 |
| Qwen2.5 1.5B Instruct | W4A16 | StackFlow | 4.63 | – | 1219.54 | source ↗ accessed 2026-10-17 |
| Llama 3.2 1B Instruct | W8A16 | StackFlow | 4.48 | – | 891 | source ↗ accessed 2026-10-17 |
| DeepSeek-R1-Distill-Qwen 1.5B | W8A16 | StackFlow | 3.57 | – | 1075.04 | source ↗ accessed 2026-10-17 |
| Qwen2.5 1.5B Instruct | W8A16 | StackFlow | 3.57 | – | 3056.54 | source ↗ accessed 2026-10-17 |
| Llama 3.2 1B | not reported | StackFlow | 3.44 | – | – | source ↗ 2026-04-24 |
| Qwen2.5 1.5B Instruct | not reported | StackFlow | 2.55 | – | – | source ↗ 2026-04-24 |
| DeepSeek-R1-Distill-Qwen 1.5B | not reported | StackFlow | 2.42 | – | – | source ↗ 2026-04-24 |
Can the M5Stack Module LLM run…
Can it run DeepSeek-R1-Distill-Qwen 1.5B?Can it run InternVL3 1B?Can it run Llama 3.2 1B?Can it run Qwen2.5 0.5B?Can it run Qwen2.5 1.5B?Can it run Qwen3 0.6B?
Have this board? Submit your own run →