OmniNeural-4B & NexaML: innovating Multimodal AI on Qualcomm Hexagon NPU

How Qualcomm and Nexa AI are driving NPU-First Innovation that unlocks natural, local, and scalable intelligence everywhere

Qualcomm Technologies, Inc. stands at the forefront of on-device AI, relentlessly pushing the boundaries of what’s possible. Every day, we witness groundbreaking advances in on-device intelligence from visionary AI companies across the globe.

Among these trailblazers is Nexa.AI, with OmniNeural-4B—a revolutionary multimodal AI model— that has been engineered from the ground up to run natively on the Hexagon NPU. This powerful synergy is ushering in a new era of on-device intelligence, empowering devices with unprecedented capabilities right at the edge.

OmniNeural-4B stands apart as the world’s first NPU-aware multi-modal model engineered to seamlessly understand and process text, voice, and vision together. Unlike conventional AI models that depend only on cloud connectivity, OmniNeural harnesses the on-device AI performance from Hexagon NPUs, ushering in a new reality where AI is both local and deeply integrated.

Whether you’re using a laptop, smartphone, automotive, XR or IoT device, this model delivers state-of-the-art AI performance right where you need it—with all the on-device AI benefits like immediacy, cost-efficiency, reliability and privacy.

Unlocking Multimodal AI Performance on Device

OmniNeural-4B is designed for high-quality natural language, vision, and audio understanding, with advanced text generation capability, while being optimized for performance on device. OEMs and Developers can deliver:

• Smarter mobile apps: Private, on-device agents that see, hear, and understand personal context—powering chat, translation, and creative tools that respond instantly without the cloud.
  
  
• Intelligent IoT devices: From smart homes to factory floors, devices reason across signals in real time to detect patterns, guide actions, and adapt naturally at the edge.
  
  
• Next-gen automotive experiences: Voice copilots that recognize drivers and surroundings, personalize cabin experiences, and assist with tasks—all directly on the vehicle’s compute platform.
  
  
• Productivity and creativity on PCs: Local AI that summarizes, creates, and automates workflows with richer context and stronger privacy—delivering results with lower latency, powered by Snapdragon X Series compute platforms

Introducing NexaML: Revolutionizing On-Device AI Inferencing

NexaML is a breakthrough solution designed for the new era of on-device AI, offering a seamless experience that combines the simplicity of cloud APIs with the power of local AI model inferencing. By running models directly on NPUs, NexaML delivers state-of-the-art performance and full compatibility with the latest advancements in AI.

Built to harness the full capabilities of the Qualcomm AI Stack, NexaML leverages Qualcomm AI Engine direct (QNN) for direct access to the Hexagon NPU, unlocking unprecedented performance from your device. With NexaML, developers can deploy sophisticated AI applications locally, benefiting from cutting-edge speed, efficiency, and support for the most advanced models available.

Here are some highlights of NexaML:

• Ease of use: One line of code after downloading the Nexa SDK, try it on your laptop powered by Snapdragon today, you can see Hexagon NPU acceleration right away
  
  
• Multimodal support: support multi-image and multi-audio inputs
  
  
• Structured decoding: Supports JSON output that enables function calling for agentic AI workflow
  
  
• Quantization: ~10% lower perplexity with NexaQuant
  
  
• Extended context length: 2× longer without speed loss
  
  
• Cross-platform: Supports PCs powered by Qualcomm technology, mobile devices, automotive platforms and IoT
  
  
• OpenAI-compatible API: for drop-in integration