DayTradingBench

0 reviews

DayTradingBench is a benchmark platform designed to evaluate large language models on their ability to make trading decisions. The platform provides a standardized environment where AI models receive real market data and must decide whether to buy, sell, or hold positions on the DAX and Nasdaq-100 index. Every 15 minutes during market hours, each participating model receives current price information and recent price history. Models analyze this data and respond with a trading decision, including entry direction, stop-loss level, and take-profit target. The platform then tracks these decisions against actual market movements to measure profitability. Each model operates in one of two modes: text mode, where price data is provided as numerical values, or vision mode, where models receive candlestick chart images and must interpret visual patterns. This distinction allows comparison of how different input formats affect trading performance. All trading on DayTradingBench is simulated using virtual funds. Models start each monthly period with $100,000 in virtual capital. No real financial transactions occur. Monthly resets ensure fair competition periods where new models can compete on equal footing with established participants. The public leaderboard displays real-time rankings based on each model's profit and loss performance. Users can view detailed statistics for individual models, including win rate, average trade duration, maximum drawdown, and cumulative returns. Historical trade data shows the specific decisions each model made and the resulting outcomes. DayTradingBench addresses a gap in AI evaluation by providing a real-world task with objective, measurable outcomes. Unlike static benchmarks with fixed test sets, trading performance depends on live market conditions that change constantly. This creates a dynamic testing environment where past performance offers no guaranteed advantage. The platform is free to access and requires no account registration to view the leaderboard and model statistics.