Six major AI models face off in a real money showdown: With an initial investment of $10,000, who will make the most money?

In an unprecedented real-money trading competition, six major AI models—including Grok, DeepSeek, Claude, and GPT—are competing to generate the highest returns from an initial $10,000 investment on the Hyperliquid perpetual contract platform.

Key details of the contest:

Each AI starts with $10,000 USD capital
Models independently analyze market data for six cryptocurrencies including BTC, ETH, and SOL
All trading decisions (orders, position closing, leverage) are made autonomously
Transactions occur on-chain via Hyperliquid for full transparency and traceability
The winner is determined by the highest final account balance

This competition aims to objectively assess the raw trading capabilities of general AI models without fine-tuning for trading specifically. The on-chain approach ensures complete transparency, avoiding potential exchange manipulation. This event highlights the growing trend of integrating AI with crypto-native protocols, potentially paving the way for AI agents to autonomously participate in DeFi, prediction markets, and other on-chain economic activities.

An unprecedented AI trading war has once again erupted in the crypto world: nof1 brought six major AI models to compete with each other in terms of trading capabilities with real money on the chain.

The rules of the game are simple and crude:

The platform collects real-time data on prices, historical charts, and other data for six mainstream cryptocurrencies, including BTC, ETH, and SOL, allowing six top-tier models, including Grok, DeepSeek, Claude, and GPT, to independently analyze market trends and execute transactions.

Each contestant starts with $10,000 USD and makes independent trading decisions such as placing orders, closing positions, and using leverage on the Hyperliquid perpetual contract platform. Whoever has the highest account balance in the end wins.

This is not just a simulation, it's a real battle!

Here are some of my personal insights:

1) nof1 AI uses Alpha Arena as a benchmark for measuring AI investment capabilities. Therefore, these are generally general AI models, not models fine-tuned for trading. This allows for an objective assessment of the trading capabilities of AI LLMs.

2) We do not employ complex "ensemble learning" methods such as voting, weighting, and multi-agent interoperability, as this would effectively become a competition of quantitative teams' strength. Arena seeks to maximize the autonomy of AI capabilities.

3) The reason for using HyperliquidX for "fair competition" in live trading, rather than connecting to an exchange through an API, undoubtedly emphasizes the need for full transparency and traceability of all transactions on-chain. This avoids uncontrollable factors in an exchange environment. This is the key to Perp Dex becoming a mainstream trading platform in the future. What if agents can replace real players in trading?

4) Public competitions of AI LLMs trading capabilities are not uncommon. I understand that many cutting-edge AI+Crypto projects are conducting similar research, such as enabling AI models to determine LP price ranges, autonomously manage DAOs, and even capture MEV opportunities. A major trend in the future is to make AI capabilities as composable as possible within crypto-native protocols such as DeFi.

5) The market has long been anticipating the emergence of killer applications in the DeFAI direction. There is still much room for imagination in allowing LLMs to participate in on-chain gaming. For example, the recently hotly debated "prediction market" is a great place for AI to train its capabilities, and there have long been many cases of agents making profits by placing bets on their own.