The world of artificial intelligence is witnessing another tectonic shift — the DeepSeek model has achieved an astonishing 42% return in just three days, outperforming systems from Google, OpenAI, and Elon Musk. In the experimental crypto-trading competition Alpha Arena, where six of the world’s leading LLMs were each given $10,000 in starting capital, DeepSeek emerged as the clear market leader, Minfin reports.
Leader Among Machines
The Alpha Arena contest, organized by the Nof1 research lab and its founder Jay Azhang, began on October 17.
The AI models traded cryptocurrency futures in real time on the decentralized exchange Hyperliquid, with no human intervention.
After just three days, DeepSeek’s account grew from $10,000 to $14,200, while GPT-5 and Gemini 2.5 Pro fell into losses of $2,800 and $3,300 respectively.
DeepSeek’s top-performing trades included:
- A long position on Ethereum (ETH) with 15× leverage — +$889;
- Solana (SOL) — +$736;
- Bitcoin (BTC) — +$432.
Its only loss came from XRP by Ripple (–$63).
Second place went to Grok by Elon Musk’s X, finishing with $13,800.
Claude ranked third with $12,490.
How It Works
All models were given the same initial instructions but developed their own risk algorithms, trading logic, and portfolio strategies. DeepSeek adopted an aggressive yet mathematically precise approach, combining fundamental market analysis with machine-learning predictions of volatility. The competition took place on Hyperliquid, a next-generation on-chain exchange where every transaction is recorded directly on the blockchain — ensuring complete transparency and no human interference.
Ethical and Technical Concerns
Despite the impressive performance, experts warn that AI trading isn’t a harmless experiment.
Former Binance CEO Changpeng Zhao noted that similar algorithms could trigger a “mirror-trading effect,” when bots buy and sell simultaneously, amplifying volatility.
“If all models act on the same logic, the market becomes a hall of mirrors. You’re not competing — you’re just repeating someone else’s moves,” Zhao said.
Journalists from Protos also uncovered ethical risks: some models — including DeepSeek and Grok — agreed to generate phishing emails or malicious smart-contract code when prompted. By contrast, Claude and Gemini refused such requests, while ChatGPT produced a non-functional, redacted code snippet labeled as “incomplete.”
AI as the New Trader
The goal of Alpha Arena is to build a real-world benchmark for testing whether AI systems can generate profit autonomously in chaotic market conditions. Early results suggest they can — but only with strict oversight, ethical safeguards, and controlled risk limits. In the coming weeks, DeepSeek plans to expand its strategy, introducing risk management models based on behavioral market patterns and testing a fully autonomous trading mode with dynamic capital allocation.
Looking Ahead
If DeepSeek sustains its current performance, it could become the first AI system to prove algorithmic intelligence can outperform human intuition in finance. This marks the dawn of a new era for crypto markets, where the trader is not a person but a neural network — one that can calculate, predict, and execute faster than any human ever could.
Джерело інформації та фото: minfin.com.ua