Skip to main content

​
Android SDK (Beta)

Nexa Android SDK NexaSDK for Android enables on-device AI inference for Android applications. It built upon Nexa AI’s proprietary NexaML engine and enables developers to run Large Language Models (LLMs), Vision-Language Models (VLMs), Embeddings, Speech Recognition (ASR), Reranking, and Computer Vision models directly on Android devices with support for NPU, GPU, and CPU inference.

​
Why NPU on mobile

​
Key Features

  • Multiple Model Types: Support for LLM, VLM, Embeddings, ASR, Reranker, and Computer Vision models
  • Hardware Acceleration: CPU, GPU, and NPU (Qualcomm Hexagon NPU) support
  • Easy Integration: Simple Kotlin/Java API with builder pattern

​
Supported Models

The following models are supported on Android ARM64 with NPU acceleration :
ModalityModel
LLMNexaAI/Qwen3-4B-Instruct-2507
NexaAI/Llama3.2-3B-NPU-Turbo
NexaAI/Granite-4-Micro
NexaAI/phi3.5-mini
NexaAI/phi4-mini
NexaAI/LFM2-1.2B-npu
VLMNexaAI/OmniNeural-4B
EmbeddingNexaAI/embeddinggemma-300m-npu
ASRNexaAI/parakeet-tdt-0.6b-v3-npu
CVNexaAI/paddleocr
NexaAI/yolo26x
NexaAI/yolo26l
NexaAI/yolo26m
NexaAI/yolo26s
NexaAI/yolo26n
NexaAI/depth-anything-v2
Beyond the NEXA-optimized models listed above, any GGUF model from the community can also run on Qualcomm Hexagon NPU. Use the cpu_gpu plugin and set device_id = "dev0", powered by the GGML Hexagon backend.

​
Get Started

Quickstart for Android

Start building with Nexa Android SDK in minutes. Follow our step-by-step guide to integrate AI models into your Android application.