Descriptions:
Sharbel A. runs a live trading experiment pitting seven major AI models against each other on Hyperliquid, a real-money crypto derivatives platform. Each contestant—Claude Opus, GPT-4o, Grok, Gemini, Kimi K2, DeepSeek, and Meta’s Llama—receives $50 and an identical prompt instructing them to trade Bitcoin, Ethereum, and Solana perpetuals with up to 20x leverage and 50% max position sizing per trade.
The results expose significant performance gaps between models. Claude Opus quickly builds its balance to $123 after round one, displaying adaptive strategy and what the creator describes as multi-move forward planning. Gemini and DeepSeek are eliminated early—the former for excessive losses despite perceived caution, the latter for high-conviction bets that backfired. Grok, the only model with live Twitter access for market sentiment, surprisingly underperforms, winning just 3 of 15 trades. Llama falls into a repetitive long-Bitcoin pattern and is eliminated in round two alongside Grok.
A custom-built live dashboard tracks every trade, running balance, and each model’s stated reasoning in real time. The experiment offers an entertaining but genuinely revealing look at how different AI architectures handle uncertainty, risk management, and strategy adaptation under pressure—with $350 in real money on the line as the ultimate performance measure.
📺 Source: Sharbel A. · Published March 04, 2026
🏷️ Format: Comparison
