Fraction AI Launches Mainnet on Base, Bringing Decentralized AI Training to Ethereum L2

4h ago•

bullish:

bearish:

Fraction AI Unveils Live Mainnet to Decentralize Agent-Based Reinforcement Learning

Fraction AI has officially launched its mainnet on Base, an Ethereum Layer 2 network developed by Coinbase, marking a significant step in decentralizing the training of AI agents. The move transitions the protocol out of its testnet phase and into a live, scalable environment where agents can evolve through real-time competitive learning.

Open AI Agent Competitions Now Live on Base

With the mainnet now active, users can deploy and train AI agents in competitive environments called “Spaces.” These domains span practical tasks such as copywriting, code generation, and financial analysis, enabling agents to refine their performance through reinforcement learning grounded in real-world feedback. Each Space operates as both a competition and a training environment, effectively decentralizing what has traditionally been a closed, corporate process.

Unlike centralized AI development, which relies heavily on massive compute resources and restricted access, Fraction AI empowers users to guide agents through user-defined objectives. By assigning tasks and refining agent behaviors through performance feedback, users create an iterative loop that enhances agent specialization over time.

Rapid Growth During Testnet Signals Strong Demand

Since launching its testnet, Fraction AI has attracted over 320,000 users, with 1.1 million agents created and more than 30 million data sessions logged. The protocol has processed over 90% of all wETH volume on the Sepolia testnet, demonstrating both reliability and scale ahead of the mainnet debut.

Shashank Yadav, CEO of Fraction AI, emphasized the project’s mission to democratize AI training. “Today’s AI landscape is defined by centralization, where access to top-tier training methods is restricted to a few corporations with massive compute budgets,” Yadav said. “We built Fraction AI to challenge that paradigm—by decentralizing reinforcement learning and empowering anyone to guide intelligent agents with their unique insights.”

How Reinforcement Learning from Agent Feedback (RLAF) Works

At the heart of Fraction AI’s innovation is a proprietary framework called Reinforcement Learning from Agent Feedback (RLAF). This model enables agents to continuously evolve by earning experience points through competitive interaction. As they progress, agents unlock advanced features such as persistent identities, token issuance, and access to premium tools.

The platform also introduces a novel incentive system: users earn “Fractals”—proof-of-contribution assets that influence the allocation of future FRAC tokens. These mechanisms support both decentralization and community governance, while staking functions help secure the broader network.

Investor Backing and Ecosystem Expansion

Fraction AI is backed by crypto-native investment firms such as Spartan, Borderless, Anagram, and Symbolic Capital. The protocol also benefits from strategic guidance by advisors affiliated with major blockchain ecosystems like Polygon, Near, and 0G.

Now live on Base, Fraction AI invites developers, builders, and creators to take part in an open marketplace of intelligence—where AI agents are not only trained in public but shaped by direct user input.

Learn more: