Product description: LLM - offline reasoning module - AX630C - for M5Core development modules - M5Stack M140
M5Stack LLM is a compact AI inference module designed for edge applications requiring intelligent offline language and voice support . With the powerful AX630C processor with built-in NPU and pre-installed Qwen2.5-0.5B , the module enables smooth handling of LLM tasks such as speech recognition, voice synthesis, and natural language analysis . It supports work with multiple models simultaneously, and its low power consumption (1.5 W at full load) makes it suitable for continuous operation. The module has a microphone , speaker, RGB LED, microSD card slot, and USB OTG connector , making it a versatile solution for AI and IoT projects.
Also check out the user guide prepared by the manufacturer!
LLM - offline reasoning module - AX630C - for M5Core development modules - M5Stack M140.
The module is equipped with the advanced SoC AX630C processor.Main features of the LLM module
- Efficient offline inference : 3.2 TOPS at INT8 precision, support for Transformer models
- Support multiple AI functions : KWS (wake word), ASR (speech recognition), LLM (language model), TTS (speech synthesis)
- Large RAM and storage : 4 GB LPDDR4 (1 GB usable) and 32 GB eMMC
- Built-in audio components : MSM421A microphone and 8 Ω 1 W speaker
- Versatile communication : UART interface and USB port with automatic master/slave switching
- Firmware update support : via SD card and USB Type-C port
- Operating status indicator : 3 RGB LEDs controlled by LP5562
- Low power consumption : 0.5W idle, 1.5W full load
Compatibility and Programmability
The LLM module is designed to be fully compatible with the M5Stack Core family , such as Core Basic , Core2 , and CoreS3 . It supports seamless plug and play integration using UART interfaces. Programming is possible both through the Arduino IDE and the UiFlow2 graphical platform, which allows easy implementation even without advanced programming knowledge. Thanks to integration with the StackFlow framework, it is possible to quickly build AI applications based on voice interaction and natural language processing.
The module comes factory-equipped with the Qwen2.5-0.5B language model.
In case of pin sharing conflicts, it is possible to cut the track on the PCB and connect it to a different set of pins using a jumper.Terminals and connection method
The M5Stack LLM module communicates via UART port, and the default transmission pins are configurable thanks to the switch pads on the PCB . You can select one of several GPIO lines depending on the compatible host (e.g. Core2, CoreS3), by cutting the path and connecting the selected pin. This avoids signal conflicts. For CoreS3, the default pins are: TX - G18, RX - G17. Power is supplied via the USB Type-C port (5V), which is also used for firmware updates and connecting peripherals such as a camera. RGB indicators inform about the working status and application updates.
Practical applications
- Offline Voice Assistants to Control Smart Home Devices
- Speech synthesis systems in interactive devices
- Educational and industrial robots with voice interface
- Control terminals in industry and logistics with AI
- AI applications without internet access where data privacy is required
The LLM module has a microphone, speaker, TF memory card, USB OTG and RGB LED. 
The module is compatible with the M5Stack Core series.
Technical specification of the module
- Processor: AX630C (Dual Cortex A53, clocked at 1.2 GHz)
- NPU: 12.8 TOPS @ INT4, 3.2 TOPS @ INT8
- RAM: 4 GB LPDDR4 (1 GB system, 3 GB acceleration)
- Flash Memory: 32GB eMMC 5.1
- Communication interface: UART, default speed 115200 bps at 8N1 (adjustable)
- Update Ports: microSD card slot / USB Type-C port
- Microphone: MSM421A
- Speaker: 8Ω at 1W
- Indicators: 3x RGB LED controlled by LP5562
- Built-in AI functions: KWS, ASR, LLM, TTS
- Power consumption: 0.5 W (idle), 1.5 W (full load)
- Operating temperature: 0 to 40°C
- Dimensions: 54 x 54 x 13 mm
- Weight: 17.1 g
Kit Contents
- 1x LLM Module
