Alpha Arena
Posted: Wed Dec 10, 2025 3:40 pm
Alpha Arena is a groundbreaking AI trading competition launched by nof1.ai, pitting top large language models (LLMs) against each other in live, autonomous trading with real capital.
What It Is and How It Works
Core Concept: Leading AI models like Grok (xAI), Qwen (Alibaba), DeepSeek, GPT (OpenAI), Gemini (Google), Claude (Anthropic), and others receive $10,000 in real money each. They trade fully autonomously—no human intervention—on live markets, analyzing data, managing risk, entering/exiting positions, and adapting in real-time.
Markets: Season 1 focused on crypto perpetual futures (e.g., BTC, ETH on Hyperliquid DEX). Season 1.5 shifted to US equities (e.g., TSLA, NVDA, MSFT, AMZN).
Transparency: All trades are on-chain, with live leaderboards tracking PNL, positions, Sharpe ratios, and more. Viewers can follow via nof1.ai or alpha-arena.org.
Goal: Benchmark LLMs' real-world trading skills in noisy, adversarial markets—revealing biases like risk appetite, hold periods, and over-trading.
Season 1 (Crypto Perps, Oct-Nov 2025)
Ran ~2 weeks; models given numerical data only (no news/tools initially).
Winner: Qwen 3 Max ($25,231 final value, ~152% gain). DeepSeek led most of it but faltered late. Others breakeven or lost (e.g., Gemini liquidated).
Key Insight: Models showed "personalities" (e.g., aggressive vs. conservative), but struggled with time-series data and fees.
Season 1.5 (US Equities, Nov 19-Dec 3, 2025)
Season 1.5 has officially concluded! The competition wrapped up on December 3, 2025, at 5:00 PM EST, after ~2 weeks of live US equities trading (Nov 19–Dec 3). Models continue running post-competition for ongoing analysis, but official results are now final. This season tested AI models on stocks like TSLA, NVDA, MSFT, AMZN, GOOGL, PLTR, and NDX, incorporating news/sentiment data, up to 20x leverage, and four parallel modes: New Baseline, Monk Mode, Situational Awareness, and Max Leverage.
What It Is and How It Works
Core Concept: Leading AI models like Grok (xAI), Qwen (Alibaba), DeepSeek, GPT (OpenAI), Gemini (Google), Claude (Anthropic), and others receive $10,000 in real money each. They trade fully autonomously—no human intervention—on live markets, analyzing data, managing risk, entering/exiting positions, and adapting in real-time.
Markets: Season 1 focused on crypto perpetual futures (e.g., BTC, ETH on Hyperliquid DEX). Season 1.5 shifted to US equities (e.g., TSLA, NVDA, MSFT, AMZN).
Transparency: All trades are on-chain, with live leaderboards tracking PNL, positions, Sharpe ratios, and more. Viewers can follow via nof1.ai or alpha-arena.org.
Goal: Benchmark LLMs' real-world trading skills in noisy, adversarial markets—revealing biases like risk appetite, hold periods, and over-trading.
Season 1 (Crypto Perps, Oct-Nov 2025)
Ran ~2 weeks; models given numerical data only (no news/tools initially).
Winner: Qwen 3 Max ($25,231 final value, ~152% gain). DeepSeek led most of it but faltered late. Others breakeven or lost (e.g., Gemini liquidated).
Key Insight: Models showed "personalities" (e.g., aggressive vs. conservative), but struggled with time-series data and fees.
Season 1.5 (US Equities, Nov 19-Dec 3, 2025)
Season 1.5 has officially concluded! The competition wrapped up on December 3, 2025, at 5:00 PM EST, after ~2 weeks of live US equities trading (Nov 19–Dec 3). Models continue running post-competition for ongoing analysis, but official results are now final. This season tested AI models on stocks like TSLA, NVDA, MSFT, AMZN, GOOGL, PLTR, and NDX, incorporating news/sentiment data, up to 20x leverage, and four parallel modes: New Baseline, Monk Mode, Situational Awareness, and Max Leverage.