Chinese AIs Best in Cryptocurrency Trading. Gemini 2.5 Pro and GPT-5 Are Deeply in the Red
One week into the Alpha Arena benchmark, where six AI models trade cryptocurrency with a $10,000 starting fund, a clear divide has emerged. While two Chinese models are showing impressive gains, their Western counterparts, including GPT-5 and Gemini 2.5 Pro, have suffered catastrophic losses.
Week One Leaderboard
- Qwen3: Capital soared to ~$17,500, a 75% increase
- DeepSeek V3.1: Showed strong performance, reaching ~$13,500
- Grok 4 & Claude Sonnet 4.5: Experienced only minor losses
- Gemini 2.5 Pro: Suffered a major setback, with capital dropping to ~$3,300
- GPT-5: Faced near-total losses, with its fund reduced to just ~$2,800
A Test of Strategy, Not Just Smarts
The competition's AIs trade based purely on technical market analysis, ignoring news and other external factors. This approach isolates each model's ability to manage risk and make strategic decisions in a highly volatile and unpredictable environment.
This benchmark is a powerful test of an AI's ability to navigate chaos. Successful crypto trading requires a steady hand and sound risk management—skills where Qwen3 and DeepSeek V3.1 are currently demonstrating clear superiority.
The Alpha Arena benchmark is expected to run for several more weeks, with a new season planned to follow. It remains to be seen if the losing models can stage a comeback or if the current leaders will maintain their dominance.