Alpha Arena AI Trading Experiment Sees China Outperform Western Models so Far

Alpha Arena Competition Shows Supremacy of Asian AI Models in Trading

The Facts:

Alpha Arena, a competition that has different artificial intelligence (AI) models trading perpetual futures on Hyperliquid, has offered interesting insights into how these models act and produce their own alpha for trading moves.

The competition, organized by financial AI laboratory Nof1, pitted six different models, including Claude 4.5 Sonnet, Deepseek V3.1 Chat, Gemini 2.5 Pro, GPT-5, Grok 4, and Qwen 3 Max, into an all-out trading battle to get the best results with a starting budget of $10,000.

The current standings after 8 days of trading action have Qwen in the lead, with $17,496.35 in its account and only one active position in BTC, followed by Deepseek.

In opposition, all other models have lost money, with GPT-5 being the worst trader so far. The OpenAI-incubated model has lost over $7,000, dropping to the bottom of the leaderboard.

Analysts have called attention to the difference between models created in China, which have been superior so far, and their Western counterparts, all of which have lost money.

Why It Is Important:

Alpha Arena seeks to measure the competence of these models in taking decisions that include real money, managing a trading portfolio. AI trading has been touted by some to be the future of trading, as AI can process high quantities and information to produce its own alpha.

Nonetheless, the experiment has shown that AI automated trading is still in the works, as results have been mixed so far, with Asian models taking the lead.

Jay Azhang, founder of Nof1, stated that this season’s goals were to determine the biases in each model, the major differences in their trading styles, and if they could follow basic risk management rules.

Looking Forward:

Nof1 has already teased a second season to test the trading skills of these models, and from their homegrown models. A human trader will also be part of the new phase.

“Overall, we’re excited by the potential of LLMs and trading, but we’re still skeptical. Much to test and learn,” Azhang concluded.

FAQ

• What is Alpha Arena?
  Alpha Arena is a competition where various AI models trade perpetual futures on Hyperliquid, examining their trading strategies and performance.

• Which AI models are competing?
  The competition includes AI models such as Claude 4.5 Sonnet, Deepseek V3.1 Chat, Gemini 2.5 Pro, GPT-5, Grok 4, and Qwen 3 Max.

• Who is leading the competition?
  Qwen currently leads with a balance of $17,496.35, while GPT-5 has underperformed, losing over $7,000.

• What are the implications of this competition?
  Alpha Arena aims to explore the efficacy of AI trading, revealing significant performance differences between Chinese and Western models, and highlighting ongoing challenges.