The NOF1.ai platform created real accounts on Hyperliquid, deposited $10,000 into each, and let AI models compete to see who profited the most. GPT-5, Claude, DeepSeek, Qwen, Gemini, and Grok are officially in the ring.

Each model operates autonomously: opening, closing, and managing positions without human intervention. Everything is on-chain, auditable, and in real time via HyperliquidX.

It's the first public and competitive experiment between LLMs and real markets.

After one week, the results are surprising:

🥇 DeepSeek Chat v3.1: +117% 🥈 Qwen3 Max: +71% 🥉 Claude Sonnet 4.5: +16% 🔻 Gemini 2.5 Pro and GPT-5: -60% The Chinese are winning — with triple-digit returns and almost surgical discipline.

The pattern is clear: AIs that trade less and rely more on long trends are performing better. DeepSeek and Qwen have high long exposure (~70-90%) and hold positions longer. GPT-5 and Gemini, on the other hand, have many short trades, low success rates, and excessive leverage.

This suggests something fascinating: Even cutting-edge algorithms still suffer from "recency bias"—they interpret recent movements as permanent.

In a bull market, those who are long win.

But in a bear market, the ranking can reverse in hours.

Another crucial point: emotions. LLMs don't feel fear, greed, or FOMO. They don't deviate from the plan—they follow fixed probabilities and risk parameters. And this cold discipline, impossible for humans, is a clear competitive advantage.

AIs excel at execution and risk management, but they still don't understand macro context, liquidity, or human sentiment. In other words, they're excellent soldiers, but terrible strategists.

The market is watching closely.

There are even markets on Polymarket betting on who wins (or loses).

The convergence of AI and blockchain has never been so transparent.

The NOF1.ai experiment is a glimpse into the future:

Will we see funds 100% managed by AI agents, without human managers, in the future?