DeepSeek Model Wins in Hong Kong and U.S. Stock Trading Competition with an Annualized Return of 10.61%, Far Exceeding GPT and Nasdaq Benchmark

The DeepSeek model, independently developed in China, recently stood out in an innovative AI trading experiment, achieving a surprising annualized return of 10.61%, surpassing global models such as GPT, Claude, and Gemini, as well as the Nasdaq 100 (QQQ) reference index that tracks technology stocks.

This open "AI-Trader" experiment conducted by the University of Hong Kong aims to subject large language models (LLMs) to a real-time trading environment in the U.S. stock market, conducting a completely human-free competition, marking a milestone in the potential for independent AI profitability in real financial markets.

AI Trading Scenario: The Key to DeepSeek's Success

During a nearly one-month trial period (until October 24), each of the five models, including DeepSeek, GPT, Claude, Gemini, and Qwen, was given $10,000 to conduct stock trades in the Nasdaq 100 index.

The rules of the experiment were extremely strict: any pre-set strategies or human suggestions were prohibited. Each model only had basic tools to check stock prices, collect news, and execute orders, making all its decisions based solely on its own algorithms and learning capabilities.

The final results showed that DeepSeek achieved a real return of 10.61%, leading the others, and its performance during this period was almost five times higher than that of the QQQ fund. This achievement clearly demonstrates the strong adaptability and practical capability of Chinese artificial intelligence in the complex and volatile U.S. stock market.

Deep Impact on Financial Technology

This advancement not only validated the practical value of LLMs in high-frequency decisions but also opened new paths for quantitative trading and algorithmic investment. The research team emphasized that the open nature of the "AI-Trader" project will help global developers further optimize AI trading systems, promoting the democratization of financial technology.

Experts point out that DeepSeek's leadership highlights China's global competitiveness in the open-source AI field. However, as AI technology advances rapidly, investors should still be cautious about market risks and algorithmic biases. For the future, the "AI-Trader" project plans to expand to more markets and models, aiming to explore the long-term role of AI in the global financial ecosystem.

Link: https://github.com/HKUDS/AI-Trader

AI Music Creation Becomes a New Side Job for Programmers: Single Track Plays Over 2 Million Times, Copyright Revenue Reaches Several Ten Thousand Yuan

In 2025, the popularity of AI music creation tools is changing the industry landscape. In January, a player from Genshin Impact used Suno to create a song with 6.4 million plays, sparking discussions about the capabilities of AI creation. Programmers have become an active group, and in March, Yapie completed a theme song using multiple tools within a few hours.

Reflection AI Raises $2 Billion to Become a Pioneer of Open-Source AI in the U.S., Challenging DeepSeek

The startup Reflection AI, founded just one year ago, has completed a $2 billion funding round, valuing the company at $8 billion, which is 15 times higher than seven months ago. The company was founded by former Google DeepMind researchers, and it has shifted from autonomous coding agents to open-source AI development, aiming to challenge closed labs like OpenAI and act as a Western counterpart to the Chinese AI company DeepSeek.

DeepSeek Gains Global Attention: How a Quantitative Trading Fund's Tech Star Is Challenging U.S. AI Supremacy

The chatbot application developed by Chinese AI lab DeepSeek has topped the download charts on Apple and Google app stores this week, sparking questions from Wall Street and the tech industry about the sustainability of the U.S. AI leadership and chip demand. The company is supported by quantitative fund Huanfang Quantitative and uses efficient computing technology to train AI models. Its founder, Liang Wenfeng, established the fund in 2015 and used AI to assist in trading decisions.

DeepSeek releases V3.2-exp model, pioneering sparse attention mechanism significantly reduces AI inference costs

DeepSeek releases the experimental model V3.2-exp, which adopts an innovative 'sparse attention' mechanism to significantly reduce the cost of long context inference. The model is now available on Hugging Face and GitHub. The core is the 'lightning indexer' and optimized attention mechanisms to improve processing efficiency. This breakthrough technology is expected to promote the development of AI in the field of long text processing.

Beware of Fraud! DeepSeek Officially Warns Users to Prevent Scams Related to Computing Power Leasing and Equity Financing

DeepSeek (Deep Seek) issued an official statement on the evening of September 18, revealing that fraudsters have been impersonating the company or its current employees, using forged employee IDs and business licenses, and engaging in fraud on multiple platforms. These scammers are posing as offering 'computing power leasing' or conducting 'equity financing' to illegally charge users. This behavior not only seriously infringes on user property rights but also harms the company's reputation. In the statement, DeepSeek emphasized that the company has never asked users to transfer money to any personal account or non-official account.