FLock Web3 Launches Decentralized AI Proxy Model

Background and Key Developments

On March 12, 2025, the decentralized AI platform FLock.io officially launched the FLock Web3 proxy model, a language model (LLM) specifically designed for on-chain tasks. Its core objective is to resolve the adaptation challenges of traditional AI in Web3 scenarios. In a Web3 function call benchmark test, the model achieved a precise matching rate of 75.93%, surpassing general models such as GPT-4o and Gemini, marking a new phase in the integration of AI and blockchain.

The uniqueness of the FLock Web3 proxy model lies in its native blockchain logic comprehension. Unlike traditional AI models that rely on extensive prompt engineering and manual fine-tuning to handle on-chain tasks, the FLock model is trained directly on on-chain data from public chains such as BSC and Ethereum through federated learning techniques. It naturally masters the logic rules in scenarios such as smart contract interactions and liquidity pool analysis. For example, the model can automatically identify arbitrage opportunities in DeFi protocols or predict market trends through cross-chain transaction data, thereby providing real-time decision support for developers and institutions.

Technical Architecture and Innovative Breakthroughs

The technical breakthroughs of the FLock Web3 proxy model are reflected in three aspects:

1. Decentralized Training Framework: Built on FLock’s mainnet AI Arena platform, developers around the world can contribute computing power and data through model training competitions and receive FLOCK token rewards. This community-driven approach has attracted over 1,500 training nodes, cumulatively creating more than 19,000 models and forming a distributed AI training network.
2. Privacy Protection Mechanism: Utilizing zero-knowledge proof (zkFL) and homomorphic encryption, the model ensures that user data is preprocessed locally and only encrypted feature parameters are shared, thus avoiding the risk of centralized data leakage. For example, when Animoca Brands developed an investment due diligence model using this technology, the original business data never left the company’s servers.
3. On-chain and Off-chain Collaboration: The model can invoke smart contract interfaces in real time and write analysis results directly to the blockchain. After integration into the IO Intelligence platform by io.net, users can automatically execute multi-chain asset allocation through natural language instructions, reducing transaction confirmation times by 60%.

In addition, the FLock model supports dynamic fine-tuning. When it detects changes in on-chain rules (such as Ethereum EIP upgrades), it can automatically trigger a retraining process to ensure that its logic is updated accordingly. This adaptive capability helps it maintain a competitive edge in the fast-iterating Web3 environment.

Market Impact and Ecosystem Collaboration

The launch of the FLock Web3 proxy model has triggered multiple market effects:

1. DeFi Efficiency Revolution: OpenGradient has used the model to optimize liquidity pool management strategies. Its ETH-USDC pool saw an annual yield increase of 22% and slippage reduced to within 0.3%.
2. Lowering Development Barriers: HashKey Chain introduced a developer assistant based on the FLock model, which can automatically generate smart contract code and detect security vulnerabilities, reducing the development cycle for beginners from three weeks to three days.
3. Unlocking Cross-chain Data Value: The model’s multi-chain analysis capabilities are being applied in RWA (Real World Asset tokenization) scenarios. For example, a real estate valuation model developed in collaboration between JP Morgan and FLock aggregates property data from chains such as Polygon and Avalanche, keeping the valuation error rate below 1.5%.

FLock’s ecosystem expansion has also benefited from capital support. In March 2024, it completed a $6 million seed round with participation from institutions such as DCG and Lightspeed Faction; in December 2024, it secured an additional $3 million strategic round led by Animoca Brands. These funds will be used to expand the federated learning node network, with a plan to cover 50% of global AI training demand by the end of 2025.

Challenges and Future Outlook

Despite its technological leadership, FLock still faces two major challenges:

1. Control of Computing Power Costs: Each training run currently consumes about 1,200 GPU hours. Although distributed computing through io.net has helped lower costs, large-scale commercial use still requires improvements in energy efficiency.
2. Regulatory Compliance: As the SEC tightens its scrutiny on AI-generated financial advice, FLock needs to collaborate with compliant exchanges (such as JuCoin) to develop KYC modules to ensure that model outputs adhere to anti-money laundering regulations.

Looking ahead, FLock plans to introduce two key upgrades. The AI Proxy Marketplace will allow developers to list customized proxy models and trade usage rights through FLOCK tokens, forming a decentralized AI application ecosystem. The Cross-chain Oracle Network upgrade, in cooperation with Chainlink, will develop a zk verification mechanism that enables the model to directly access real-world data (such as Federal Reserve interest rate decisions), expanding its application to macroeconomic analysis.

The FLock Web3 proxy model is not only a technological breakthrough but also a key milestone in the democratization of AI. Through community collaboration, privacy protection, and native on-chain adaptation, it is redefining the smart infrastructure for the Web3 era.