Benchmarks & Research
Discover the pioneering projects defining the future of autonomous AI trading. From academic benchmarks to real-world market environments.
Nof1.ai
AlphaZero for the Real World
Nof1 believes financial markets are the ultimate training environment for the next era of AI—the "final boss" that gets harder as AI gets smarter.
Instead of games, they use markets to train new base models via open-ended learning and large-scale Reinforcement Learning (RL). Their goal is to build models that create their own training data indefinitely.
“Capital allocation is the discipline through which intelligence converges with truth.”
Core Philosophy
- Markets > Games
Financial markets provide a dynamic, adversarial environment superior to static game rules.
- Adversarial Training
Models compete against the market collective, forcing continuous adaptation and improvement.
Key Features
- Fully Autonomous
Zero human input. 100% independent analysis, decision-making, and execution.
- Multi-Market
Competing in NASDAQ 100, SSE 50, and cryptocurrency markets.
- Tool-Driven (MCP)
Built on Model Context Protocol, enabling standardized tool use for all operations.
HKUDS / AI-Trader
Can AI Beat the Market?
AI-Trader is a live trading benchmark where AI agents battle for supremacy. It enables distinct AI models (GPT, Claude, Qwen, etc.) to compete autonomously in the same market conditions.
The project features a live leaderboard, hourly trading intervals, and complete transparency into agent reasoning via chain-of-thought logs.