DeepSeek Surpasses GPT-5 in Cryptocurrency Trading Competition

DeepSeek V3.1 achieved a 120% return over nine days in the Alpha Arena competition.

The Chinese AI model DeepSeek V3.1 achieved a 120% return over nine days in the Alpha Arena cryptocurrency trading competition hosted by nof1 laboratory. It earned $22,031 from an initial capital of $10,000.

The competition will continue until November 3. Six AI models were each given $10,000, identical instructions, and market data. Their task is to maximize profits by trading Bitcoin, Ethereum, and Dogecoin.

The Qwen 3 Max neural network from Alibaba Group holds the second interim position with an 80% return, increasing its balance to $18,014. Qwen briefly led DeepSeek but then relinquished the top spot.

Models from OpenAI and Google DeepMind suffered significant losses. GPT-5 lost 61.5% of its capital, leaving $3,845. Gemini 2.5 Pro showed a similar result with a 59% loss ($4,077).

Other participants demonstrated more modest results. Grok 4 from xAI earned 9% ($10,911), while Claude 4.5 Sonnet from Anthropic achieved 21% ($12,122).

Earlier, DeepSeek introduced a new multimodal AI capable of processing large and complex documents using significantly fewer tokens.