Alibaba’s Qwen large language model has taken first place in the “Alpha Arena” real-time AI investment competition, achieving a 22.32% return over 17 days. The contest, launched on October 18 by research firm Nof1, featured six leading global AI models — Qwen3-Max, DeepSeek v3.1, GPT-5, Gemini 2.5 Pro, Claude Sonnet 4.5, and Grok 4 — each given $10,000 in initial capital to autonomously trade in live markets with no human intervention. Alibaba’s Qwen and China’s DeepSeek were the only two models to end the competition with profits, while all four U.S.-based models posted losses. OpenAI’s GPT-5 performed the worst, recording a loss of more than 62%, according to the final results.