Fetch.ai launches first Web3 agentic AI model

Fetch.ai has launched ASI-1 Mini, a native Web3 large language model designed to support complex agentic AI workflows.

Described as a gamechanger for AI accessibility and performance, ASI-1 Mini is heralded for delivering results on par with leading LLMs but at significantly reduced hardware costs—a leap forward in making AI enterprise-ready.

ASI-1 Mini integrates into Web3 ecosystems, enabling secure and autonomous AI interactions. Its release sets the foundation for broader innovation within the AI sector—including the imminent launch of the Cortex suite, which will further enhance the use of large language models and generalised intelligence.

“This launch marks the beginning of ASI-1 Mini’s rollout and a new era of community-owned AI. By decentralising AI’s value chain, we’re empowering the Web3 community to invest in, train, and own foundational AI models,” said Humayun Sheikh, CEO of Fetch.ai and Chairman of the Artificial Superintelligence Alliance.

“We’ll soon introduce advanced agentic tool integration, multi-modal capabilities, and deeper Web3 synergy to enhance ASI-1 Mini’s automation capabilities while keeping AI’s value creation in the hands of its contributors.”

Democratising AI with Web3: Decentralised ownership and shared value  

Key to Fetch.ai’s vision is the democratisation of foundational AI models, allowing the Web3 community to not just use, but also train and own proprietary LLMs like ASI-1 Mini. 

This decentralisation unlocks opportunities for individuals to directly benefit from the economic growth of cutting-edge AI models, which could achieve multi-billion-dollar valuations.  

Through Fetch.ai’s platform, users can invest in curated AI model collections, contribute to their development, and share in generated revenues. For the first time, decentralisation is driving AI model ownership—ensuring financial benefits are more equitably distributed.

Advanced reasoning and tailored performance  

ASI-1 Mini introduces adaptability in decision-making with four dynamic reasoning modes: Multi-Step, Complete, Optimised, and Short Reasoning. This flexibility allows it to balance depth and precision based on the specific task at hand.  

Whether performing intricate, multi-layered problem-solving or delivering concise, actionable insights, ASI-1 Mini adapts dynamically for maximum efficiency. Its Mixture of Models (MoM) and Mixture of Agents (MoA) frameworks further enhance this versatility.  

Mixture of Models (MoM):  

ASI-1 Mini selects relevant models dynamically from a suite of specialised AI models, which are optimised for specific tasks or datasets. This ensures high efficiency and scalability, especially for multi-modal AI and federated learning.  

Mixture of Agents (MoA):  

Independent agents with unique knowledge and reasoning capabilities work collaboratively to solve complex tasks. The system’s coordination mechanism ensures efficient task distribution, paving the way for decentralised AI models that thrive in dynamic, multi-agent systems.  

This sophisticated architecture is built on three interacting layers:  

Foundational layer: ASI-1 Mini serves as the core intelligence and orchestration hub.  

Specialisation layer (MoM Marketplace): Houses diverse expert models, accessible through the ASI platform.  

Action layer (AgentVerse): Features agents capable of managing live databases, integrating APIs, facilitating decentralised workflows, and more.  

By selectively activating only necessary models and agents, the system ensures performance, precision, and scalability in real-time tasks.  

Transforming AI efficiency and accessibility

Unlike traditional LLMs, which come with high computational overheads, ASI-1 Mini is optimised for enterprise-grade performance on just two GPUs, reducing hardware costs by a remarkable eightfold. For businesses, this means reduced infrastructure costs and increased scalability, breaking down financial barriers to high-performance AI integration.  

On benchmark tests like Massive Multitask Language Understanding (MMLU), ASI-1 Mini matches or surpasses leading LLMs in specialised domains such as medicine, history, business, and logical reasoning.  

Rolling out in two phases, ASI-1 Mini will soon process vastly larger datasets with upcoming context window expansions:  

Up to 1 million tokens: Allows the model to analyse complex documents or technical manuals.

Up to 10 million tokens: Enables high-stakes applications like legal record review, financial analysis, and enterprise-scale datasets.  

These enhancements will make ASI-1 Mini invaluable for complex and multi-layered tasks.  

Tackling the “black-box” problem  

The AI industry has long faced the challenge of addressing the black-box problem, where deep learning models reach conclusions without clear explanations.

ASI-1 Mini mitigates this issue with continuous multi-step reasoning, facilitating real-time corrections and optimised decision-making. While it doesn’t entirely eliminate opacity, ASI-1 provides more explainable outputs—critical for industries like healthcare and finance.  

Its multi-expert model architecture not only ensures transparency but also optimises complex workflows across diverse sectors. From managing databases to executing real-time business logic, ASI-1 outperforms traditional models in both speed and reliability.  

AgentVerse integration: Building the agentic AI economy

ASI-1 Mini is set to connect with AgentVerse, Fetch.ai’s agent marketplace, providing users with the tools to build and deploy autonomous agents capable of real-world task execution via simple language commands. For example, users could automate trip planning, restaurant reservations, or financial transactions through “micro-agents” hosted on the platform.

This ecosystem enables open-source AI customisation and monetisation, creating an “agentic economy” where developers and businesses thrive symbiotically. Developers can monetise micro-agents, while users gain seamless access to tailored AI solutions.  

As its agentic ecosystem matures, ASI-1 Mini aims to evolve into a multi-modal powerhouse capable of processing structured text, images, and complex datasets with context-aware decision-making.  

See also: Endor Labs: AI transparency vs ‘open-washing’

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.