Alibaba AI Wins Crypto Contest While ChatGPT Flops

The AI Trading Gauntlet

In a fascinating face-off between artificial intelligence and the volatile world of cryptocurrency, a recent competition put six leading large language models (LLMs) to the test. Organized by Nof1, the "Alpha Arena" contest gave each AI a $10,000 starting balance and tasked them with trading digital assets over two weeks using identical prompts and limited data. The results were a stark reminder that even the most advanced AI can struggle with market realities.

A Tale of Two AIs Winners and Losers

The final leaderboard revealed a dramatic split in performance. OpenAI's ChatGPT, one of the most recognized names in AI, finished in last place, losing a staggering $6,267, or 63% of its initial funds. Other well-known models also ended deep in the red:

• Google's Gemini: Down $5,671
• X's Grok: Down $4,531
• Anthropic's Claude Sonnet: Down $3,081

However, not all AI traders fumbled. Two models from China bucked the trend and ended with a profit. Alibaba's Qwen3 Max claimed the top spot with a respectable $2,232 gain, while DeepSeek secured second place with a $489 profit. This divergence highlights that not all AI models approach financial risk in the same way.

Why Most AIs Failed The Trading Test

According to the event organizer Nof1, a major factor in the poor performance was high trading costs. In the early stages of the competition, many of the AI systems engaged in over-trading. They executed a high volume of trades for small gains, which were quickly wiped out by fees. For instance, Alphabet's Gemini model made 238 trades, while Claude made only 38. Across the board, win rates for all models hovered between a modest 25% and 30%.

The Secret to AI Trading Success

So, how did Alibaba's Qwen3 Max manage to succeed where others failed? The key wasn't a higher win rate but a more disciplined strategy. Despite incurring the highest total fees at $1,654, its careful trade selection and avoidance of over-trading allowed it to remain consistently profitable. This performance stood in sharp contrast to ChatGPT's heavy losses, showcasing a crucial difference in risk management and execution under identical conditions.

Nof1 founder Jay Azhang called the event a "controlled stress test" for these generative AI systems. He noted that LLMs are not inherently skilled at handling the numerical time-series data common in financial markets. Interestingly, Azhang observed that each AI developed a unique "investing personality," suggesting their market behaviors could become predictable over time.

Key Takeaways for Investors

The results of the Alpha Arena offer several important lessons. Firstly, it demonstrates that an AI's ability to sound confident does not translate to success when real financial risk is involved. Though all LLMs worked from the same data, their outcomes varied widely, much like human traders with different habits and risk tolerances.

Qwen3 Max's victory underscores that in trading, discipline often beats pure predictive power. Success came from avoiding costly mistakes, not from making more trades. For investors exploring AI tools, this contest is a powerful reminder that artificial intelligence can be a valuable aid for market analysis, but it cannot replace a well-thought-out strategy and robust risk management.