The rise of Large Language Models (LLMs) has changed how we write, code, and search. But can AI navigate the high-stakes, volatile world of cryptocurrency trading better than humans? A groundbreaking, real-money competition called Alpha Arena is currently testing this question, giving six top AI models US$10,000 each in starting capital to trade autonomously.
The results, according to the latest updates, are anything but predictable, featuring staggering triple-digit gains for the leaders and near-total account wipeouts for some of the world’s most famous AI platforms. The outcome of this contest, which runs until November 3, 2025, promises to offer valuable insights into the future of AI-driven finance.
The Ultimate Intelligence Test: Alpha Arena Setup
Alpha Arena was launched by the US research firm Nof1 (or nof1.ai) with a clear mission: to create benchmarks that are closer to the real world, noting that markets are “perfect for this”. The ultimate test of intelligence, according to the project, will take place in the market.
The rules are simple but brutal: six leading AI models were deployed, each receiving US$10,000 to trade independently and autonomously. The competition began around October 17 or 18, 2025.
The AIs’ objective is to maximize returns by trading cryptocurrencies, including bitcoin, ether, and dogecoin. Unlike static simulations, the models trade 24/7 in real market conditions on the Hyperliquid decentralized exchange (DEX), utilizing perpetual futures contracts with leverage to maximize opportunities.
The six competitors include models from global tech giants and rising stars:
- DeepSeek Chat V3.1
- Qwen 3 Max (Alibaba Group Holding)
- GPT-5 (OpenAI)
- Gemini 2.5 Pro (Google DeepMind)
- Grok 4 (xAI)
- Claude 4.5 Sonnet (Anthropic)
You may be interested to Read this: Gemini vs ChatGPT vs Perplexity vs 3 more AI models: Everything Compared
The entire process is fully transparent, with all wallet addresses, transactions, and performance metrics published on-chain for public verification.
The Leaders: Chinese Models Dominate the Crypto Charts
As of an update on a recent Monday afternoon, two models had achieved gains that more than doubled their initial capital, widening their lead as they adjusted their strategies amid volatile market conditions. Both leading models are associated with Chinese platforms:
- DeepSeek Chat V3.1: DeepSeek is currently the leader of the pack. It grew its initial US$10,000 to US$22,500, securing an impressive 125 per cent gain since trading began. At one point, after 72 hours of live trading, DeepSeek Chat V3.1 led the leaderboard with a total account value of $13,830.
- Strategy Insight: DeepSeek’s strategy has been to hold all profitable long positions across assets like Ethereum, Solana, XRP, BTC, DOGE, and BNB, noting that none of its invalidation conditions had been triggered, allowing the gains to ride.
- Qwen 3 Max (Alibaba): Developed by Alibaba Group Holding, Qwen closely follows the leader. It posted a 95 per cent gain, nearly doubling its account value to US$19,600. Qwen even briefly exceeded 100 per cent returns and managed to overtake DeepSeek between October 24 and 26.
- Strategy Insight: On a recent Monday morning update, Qwen 3 Max was observed to have gone “all in” on ether (Ethereum).
The strong portfolio gains by both DeepSeek and Qwen coincided with a market rebound, specifically bitcoin’s return to about US$114,000 following mid-October volatility, with ether also advancing steadily.
The Underperformers: Big Tech Struggles
While the Chinese models raced ahead, some of the most famous Western LLMs have struggled significantly, losing substantial portions of their starting capital.
- OpenAI’s GPT-5: This model has been the worst performer so far in the competition. GPT-5 lost nearly 60 per cent of its capital, leaving it with only around US$4,000 remaining. Earlier in the competition, it was reported to have lost $7,200 in just one week. During that period, 42 of its 44 transactions ended in a loss. Despite the severe losses, the system continues to trade, attempting to “win back” its previous losses. However, recent data suggests GPT-5 has been adapting, extending a winning streak to seven days using disciplined Bitcoin (BTC) laddering.
- Google DeepMind’s Gemini 2.5 Pro: This model showed a similar result, posting a 57 per cent loss. Recently, Gemini 2.5 Pro rotated its positions into ETH/SOL pairs, leveraging reinforcement learning driven volatility filters to attempt to lock in gains.
Other models showed modest gains: xAI’s Grok 4 posted returns of 13 per cent, while Anthropic’s Claude 4.5 Sonnet posted 24 per cent returns. Claude Sonnet 4.5 remains defensive, focusing on basis trades and keeping leverage below 5x across the board.
What This Competition Really Proves
The Alpha Arena competition is more than just a leaderboard; it is a critical benchmark for measuring the investing abilities of AI in a scenario that mirrors the complexities and speed of a real market.
If AI models consistently demonstrate the ability to generate “alpha” (market-beating returns) in real conditions, it could fundamentally change finance. AI-assisted trading is already expected to increase overall market trading volumes.
Perhaps the most significant long-term consequence is the potential for Copy AI trading. If AIs consistently prove profitable over multiple seasons, people may start copying AIs in the same way they currently copy successful human traders. The distinct advantage of the AI models is their lack of emotion. Unlike human traders who might cut profits short or hold onto losing positions out of fear or hope, AIs follow their strategy consistently.
This real-time, transparent testing (with performance metrics like PNL and Sharpe ratios tracked publicly) is demonstrating which platforms possess the intelligence—or the luck—to thrive under the intense pressure of financial markets. For now, the Chinese models are showing that AI has the potential to turn a $10,000 stake into a fortune.
With some AI models generating 125% returns and others losing 60% of their $10,000 capital in the Alpha Arena, the results are wildly inconsistent. Given this high-stakes volatility, would you trust an AI model to trade your money, fully based on its autonomous decisions?
Drop your thoughts in comments below!
Key Takeaways
- Alpha Arena is testing AI’s ability to trade cryptocurrency autonomously with real money.
- Early results show significant performance differences between AI models, with Chinese models currently leading.
- The competition highlights the potential for AI-driven finance and copy AI trading.
Join our community by subscribing to our Weekly Newsletter to stay updated on the latest AI updates and technologies, including the tips and how-to guides.
(Also, follow us on Instagram (@inner_detail) for more updates in your feed and our WhatsApp Channel to get daily news straight to your Messaging App).
