The M5Stack Module LLM is yet another box-shaped device from the company that provides artificially intelligent control without internet access. It is described as an “integrated offline Large Language Model (LLM) inference module” which can be used to implement local LLM-based solutions in smart homes, voice assistants, and industrial control.
Module LLM is powered by the AX630C SoC, equipped with 4GB LPDDR4 memory, 32GB storage, and a 3.2 TOPS (INT8) or 12.8 TOPS (INT4) NPU. M5Stack says the main chip has an average runtime power consumption of 1.5W, making it suitable for long-term operation. It has a built-in microphone, speaker, microSD card slot, and USB OTG. The USB port can connect peripherals such as cameras and debuggers, and the microSD card slot supports cold and hot firmware updates.
The M5Stack Module LLM joins the list of other offline, on-device LLM-based solutions, such as the SenseCAP Watcher, Useful Sensors’ AI in a box, and Radxa Fogwise Airbox. It is compatible with the CoreMP135, CoreS3, and Core2 IoT controllers.
M5Stack Module LLM specifications:
- SoC – Axera Tech AX630C
- CPU – Dual-core Arm Cortex-A53 @ 1.2 GHz; 32KB I-Cache, 32KB D-Cache, 256KB L2 Cache
- NPU – 12.8 TOPS @ INT4 (max), 3.2 TOPS @ INT8
- ISP – 4K @ 30fps
- Video – Encoding: 4K; Decoding:1080p @60fps, H.264 only
- Supports single-channel RGMII / RMII PHY interface
- Memory – 4GB LPDDR4 RAM (1GB system + 3GB dedicated for hardware acceleration)
- Storage – 32GB eMMC 5.1 flash, microSD card slot
- Audio
- Audio Driver: AW8737
- Speaker: 8Ω @ 1W, 2014 cavity speaker
- Built-in microphone
- AI audio features – Text-to-speech (TTS), Automatic Speech Recognition (ASR), Keyword Spotting (KWS)
- USB – 1x USB OTG port
- Serial – 1x UART (115200 @ 8N1 default baud rate)
- Expansion – 8-pin FPC interface for connecting Ethernet debugging kit (under speaker)
- Misc
- 3x RGB LED (status indication)
- 1x Boot button
- Power Supply – 5V via USB-C port
- Power Consumption
- Idle: 5V @ 0.5W
- Full Load: 5V @ 1.5W
- Dimensions – 54 x 54 x 13 mm
- Weight – 17.4g
- Operating Temperature – 0 to 40°C
The Module LLM is integrated with the StackFlow framework and is compatible with Arduino and UIFlow libraries. Tutorials and other information are available on the M5Stack’s documentation website. The device compares favorably with the Jetson Orin Nano @ 10 Watts as shown in the comparison chart below.
It is compatible with various models and comes with the Qwen2.5-0.5B language model pre-installed. This model provides wake-word, text-to-speech, and speech recognition support for standalone operation and pipeline systems. M5Stack says the module will support the Qwen2.5-1.5B, Llama3.2-1B, and InternVL2-1B models in the future. It supports the computer vision models, CLIP and YoloWorld, with planned updates for DepthAnything, SegmentAnything, and other models.
The Module LLM is $49.90 on M5Stack’s store but was out of stock at the time of writing. The LLM debugging kit is sold separately (also not available) and can be used to add a 100 Mbps Ethernet port and a kernel serial port to the module for use as a single-board computer. The Module LLM and debugging should eventually become available on M5Stack’s Amazon and AliExpress stores.
Tomisin is a writer specializing in hardware product reviews, comparisons, and explainers. He is very passionate about small form factor and single-board computers.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
