We are building small but powerful on-device multimodal models for super AI agents --- fast, accurate, energy-efficient
Empower billions of devices — smartphones, automobiles, robots, laptops, smart home devices, AR/VR, and more — with fast, secure, accurate, and energy-efficient on-device AI agents
smallest, most powerful on-device multimodal model for super AI agents
Less than 1B parameters
Processes both text and images for function calling
On par with a combination of GPT-4V and GPT-4
Fluent in English and Mandarin
Demos for VR/AR and other devices are coming soon.
Nexa AI envisions a world where everyone has access to secure, eco-friendly, and ultra-responsive AI——transforming tasks into immediate, actionable solutions right at your fingertips
2 billion parameter language model developed for high-performance function calling on edge devices.
Uses a novel "functional token" strategy to reduced context length by 95%.
Demonstrated 36x faster inference speed compared to RAG solution, and 168% faster than GPT-4-turbo.
Achieved 98%+ function call accuracy, surpassing previous models and matching the performance of GPT-4.
“a groundbreaking new framework for on-device AI agents. The new era of on-device AI agents is coming.”
“Extremely fast, better than Llama+RAG, great results”
“These models possess the crucial ability to call functions, which is essential in creating”
“Interesting idea to incorporate the functions into the model with fine-tuning to get reliable generation from small LLMs.”
“For all things tech, Techware is my ultimate destination. Quality, range, and service—impeccable.”
“With just 100 training samples, the model achieved 98% accuracy in selecting the right function, surpassing GPT-4.”
“a groundbreaking new framework for on-device AI agents. The new era of on-device AI agents is coming.”
“Extremely fast, better than Llama+RAG, great results”
“These models possess the crucial ability to call functions, which is essential in creating”
“Interesting idea to incorporate the functions into the model with fine-tuning to get reliable generation from small LLMs.”
“For all things tech, Techware is my ultimate destination. Quality, range, and service—impeccable.”
“With just 100 training samples, the model achieved 98% accuracy in selecting the right function, surpassing GPT-4.”
“a groundbreaking new framework for on-device AI agents.”
“A monumental leap in function calling efficiency on devices, making real-world applications faster and smarter than ever imagined.”
“Octopus v2 represents a major leap towards making powerful AI accessible to everyone.”
“Octopus v2 showcases the potential to revolutionize how we interact with technology, emphasizing efficiency and privacy.”
“A Novel method enabling on-device models with 2 billion parameters to outperform GPT-4 in accuracy and latency, reducing context length by 95%.”
“I envisioned this last year and it’s happening. Super interesting. So fast.”
“This is amazing and will pave the path for agents on edge devices.”
“This 2B LLMs is a breakthrough in the application of LLMs for function calling, specifically tailored for Android APIs.”
“A monumental leap in function calling efficiency on devices, making real-world applications faster and smarter than ever imagined.”
“Octopus v2 represents a major leap towards making powerful AI accessible to everyone.”
“Octopus v2 showcases the potential to revolutionize how we interact with technology, emphasizing efficiency and privacy.”
“A Novel method enabling on-device models with 2 billion parameters to outperform GPT-4 in accuracy and latency, reducing context length by 95%.”
“I envisioned this last year and it’s happening. Super interesting. So fast.”
“This is amazing and will pave the path for agents on edge devices.”
Explore our collection of 200+ Premium Webflow Templates
