The six major AI models are truly competing on-chain, and in just three days, some people made +25%, while others lost -40%!

The most outrageous thing is: using the same prompts and looking at the same candlestick charts, DeepSeek crushes GPT-5 and Gemini. Current record:

🥇 DeepSeek Chat: +25.33%（$12533）
🥈 Grok-4: +21.47% ($12147)
🥉 Claude Sonnet: +10.47% ($11047)
📉 GPT-5：-25.58%（$7442）
📉 Gemini 2.5: -39.38% ($6062)

Gemin made 44 transactions, averaging 15 per day, living like a frightened bird. Claude only made 3 transactions, steady as an institution. The key point is — all of them used $10,000 in real money to trade BTC, ETH, SOL, and other coins on Hyperliquid, not on a demo account!

The most interesting thing is that DeepSeek is backed by a quantitative fund, which has been involved in algorithmic trading for decades and is now returning to its roots, making it the strongest. In contrast, Google's and OpenAI's models may rely too heavily on NLP and do not understand structured market data well enough.

Some people have started to bottom-feed on DeepSeek's orders, while others are specifically doing the opposite on Gemini. But this is awkward—once everyone knows what AI will buy, can this strategy still be used? This is the paradox of "the observer effect."

The essence of this wave of testing is actually asking: In a results-oriented crypto market, AI that can make money is more valuable than AI that can chat.