February 25, 2025 – Fetch.ai Inc., a founding member of the Artificial Superintelligence Alliance, has launched ASI-1 Mini, marking the debut of the first Web3-native large language model (LLM), specifically built and optimized for supporting complex agentic workflows. Benchmarking results demonstrate that ASI-1 Mini delivers performance on par with leading LLMs currently available in the market with significantly lower hardware costs. With seamless Web3 integration, ASI-1 Mini will enable users to interact with AI securely and autonomously.
Powered by $FET through the ASI wallet integration, ASI-1 Mini is the first model in the ASI: family, an ASI Alliance initiative set to expand with the upcoming launch of the Cortex group of models - centered around use of large language models and generalized intelligence. This pioneering approach democratizes access to foundational AI models, allowing the Web3 community to invest, train and own them. By doing so, it enables individuals to directly benefit from the growth of these models that could unlock multi-billion-dollar valuations, creating a new era of decentralized ownership and shared value in AI technology.
ASI-1 Mini introduces next-level adaptive reasoning, context-aware decision-making. It features native reasoning support with four dynamic reasoning modes, intelligently selecting from Multi-Step, Complete, Optimized, and Short Reasoning, balancing depth, efficiency, and precision. Whether tackling complex, multi-layered problems or delivering concise, high-impact insights, ASI-1 Mini ensures reasoning is always tailored to the task at hand.
"ASI-1 Mini is the first major product from the ASI Alliance’s innovation stack, marking the beginning of the ASI: rollout and a new era of community-owned AI. This launch sets the foundation for a decentralized ecosystem where the Web3 community can invest, train, and directly benefit from cutting-edge AI models,” said Humayun Sheikh, CEO of Fetch.ai and chairman of the ASI Alliance. “ASI-1 Mini is just the start—over the coming days, we will be rolling out advanced agentic tool-calling, expanded multi-modal capabilities, and deeper Web3 integrations. With these enhancements, ASI-1 Mini will drive agentic automation while ensuring that AI’s value creation remains in the hands of those who fuel its growth."
ASI-1 Mini introduces a next-generation AI architecture by extending the Mixture of Experts (MoE) framework into a Mixture of Models (MoM) and Mixture of Agents (MoA) approach. This enables a more decentralized, efficient, and scalable system, optimizing speed, resource allocation, and autonomous decision-making across diverse tasks.
Mixture of Models (MoM): Instead of a monolithic structure, ASI-1 Mini dynamically selects from multiple specialized models, each optimized for specific tasks or data types. A gating mechanism ensures that only the most relevant models are activated, enhancing efficiency, speed, and scalability. This is particularly beneficial for multi-modal AI, federated learning, and task-specific pipelines.
Mixture of Agents (MoA): Autonomous agents, each with independent reasoning, knowledge, and decision-making, collaborate seamlessly to solve complex tasks. A coordination mechanism ensures efficient task distribution, making the system resilient, decentralized, and adaptive. MoA is critical for multi-agent systems, decentralized AI, and collaborative intelligence in dynamic environments. Agents act as the intelligent I/O and execution components of the architecture.
The architecture envisions a three-layered system:
Foundational Layer (ASI-1 Mini): This layer acts as the central intelligence and orchestration point. ASI-1 Mini, with its MoE architecture and optimization for agent/tool calling
Specialization Layer (MoM Marketplace): This layer houses a collection of AI models (MoMs) with various specializations, created and offered through the ASI: platform. Each MoM is designed for a specific domain or task, providing expert-level inference within its area of specialization.
Action Layer (Agents on Agentverse): This layer consists of a variety of agents, each with specific capabilities:
These layers interact to enable complex, multi-step tasks by combining the reasoning power of ASI-1 Mini with the specialized knowledge offered by MoMs and the execution capabilities of diverse agents.
By integrating MoM and MoA, ASI-1 Mini achieves unparalleled adaptability and efficiency, activating only the necessary models and agents for any given task—ensuring optimal performance with precision, speed, and scalability.
ASI-1 Mini revolutionizes AI efficiency by delivering high-performance execution with significantly lower hardware requirements. ASI-1 Mini operates seamlessly on just two GPUs, making enterprise-grade AI more accessible and cost-effective. This results in 8x greater hardware efficiency, reduced infrastructure costs, and increased scalability—allowing businesses to integrate AI without prohibitive investment.
ASI-1 Mini is a highly versatile performer on Massive Multitask Language Understanding (MMLU) benchmarks, often matching or surpassing industry leaders in domain-specific tasks. It excels in medical sciences, history, logical reasoning, and business applications, consistently competing with top LLMs across multiple domains. This strong foundation establishes ASI-1 Mini as a well-rounded model capable of excelling in high-stakes decision-making, research-intensive fields, and enterprise-level AI applications.
ASI-1 Mini will soon process exponentially larger amounts of information through its expanded context window, rolling out in two phases:
These advancements will unlock new capabilities in decision-making and automation, making AI more effective for high-stakes applications.
The black-box problem refers to situations where AI systems, particularly deep learning models, generate outputs without offering understandable explanations for how they arrived at those conclusions. For example, a healthcare AI tool might predict the risk of a certain disease but fail to clarify which factors influenced its decision-making process.
Unlike models that reason only at the start of a task, ASI-1 takes a step further in addressing the black-box problem by employing continuous multi-step reasoning. This allows for real-time corrections, optimized decision-making, and greater reliability. While it doesn't entirely eliminate the opacity, ASI-1 significantly enhances transparency by providing more explainable outputs—critical for high-stakes applications in healthcare, finance, and other industries.
Through its multi-expert model architecture, ASI-1 fosters intelligent collaboration among specialized AI agents, optimizing complex workflows across diverse sectors. Whether managing live databases, integrating external APIs, or overseeing real-time business operations, ASI-1 outperforms traditional models in both speed and accuracy, providing clearer insights into how decisions are made.
ASI-1 Mini, part of the Cortex collection, ushers in a new era of community-driven AI innovation. With ASI:, the Web3 community can now directly participate in the training and development of advanced AI models. By decentralizing the training process, ASI: empowers individuals to contribute to the creation of cutting-edge LLM and specialist modelswithin the Cortex collection, reshaping how AI is developed and owned.
This collaborative ecosystem allows the community to stake, train, and own AI models, ensuring that the financial rewards from AI-driven advancements are more equitably distributed. By providing access to curated model collections, ASI: enables users to not only invest in but also share in the revenue generated by these transformative technologies.
With an emphasis on community contributions, ASI: accelerates innovation through its decentralized compute network (ASI Compute), creating opportunities for members to refine and build AI models together. This approach fosters shared resources, democratizing AI development, and driving solutions that revolutionize industries while enabling new business models.
ASI-1 will redefine agentic AI with unmatched capabilities:
ASI-1 Mini will be connected to AgentVerse, Fetch.ai’s agent marketplace, enabling users to build and deploy autonomous agents capable of executing real-world tasks with simple language commands. Once fully integrated, ASI-1 Mini will empower users to request complex tasks—such as booking hotels or dinner reservations—using plain language. These requests will be compiled into micro-agents, hosted within AgentVerse, that seamlessly execute tasks. Developers can monetize these micro-agents, pioneering a new open-source ecosystem where customized AI functionality is easily accessible, and the agentic economy thrives.
ASI-1 Mini is available today as a tiered freemium product for $FET holders, giving users immediate access to its powerful capabilities. In the coming weeks, agentic tool-calling capabilities will be rolled out, enabling even more advanced functionality. Additionally, users can connect their Web3 wallets for a seamless and personalized experience.
With ASI-1 Mini, Fetch.ai is delivering the future of intelligent, autonomous AI—ready to transform businesses and Web3 ecosystems alike.
ASI-1 Mini is built for multi-modal understanding and will soon expand its ability to process and generate insights across multiple data types. As its agentic workflows evolve, ASI-1 Mini will achieve more precise and context-aware decision-making, ensuring seamless adaptability across structured text, images, and complex datasets.
Fetch.ai inc, a Delaware AI company and founding member of ASI Alliance, is redefining the possibilities of an intelligent and connected world through its AI agent-based technology. Built on Cosmos, Fetch.ai's infrastructure technology enables developers and businesses to build, deploy & monetize through an agent-based modular platform for the new generation of AI applications. The company's core product, Agentverse, fuses Language Models (LLMs) and AI Agents to create an open and dynamic marketplace that connects users to services and reimagines the current search experience. For additional information visit: fetch.ai
The Artificial Super Intelligence (ASI) Alliance is a collective formed by Fetch.ai, SingularityNET (SNET), Ocean Protocol and CUDOS. As the largest open-sourced, independent entity in AI research and development, this alliance aims to accelerate the advancement of decentralized Artificial General Intelligence (AGI) and, ultimately, Artificial Superintelligence (ASI). For additional information on ASI, visit: superintelligence.io
