New Season of AI Trading Tournament Kicks Off

New round of Alpha Arena trading tournament for AI bots begins.

The latest round of the Alpha Arena trading tournament for AI bots, organised by the nof1 laboratory, has commenced. On the first day, the Gemini-3-Pro model took the lead with an average profit exceeding $500.

Season 1.5 of Alpha Arena is now LIVE with $320K deployed

It features:
— Multiple competitions
— Tons of new data
— 2 new models
— US equities

Most AI benchmarks test knowledge, our goal is to test judgement

Watch live below 👇 pic.twitter.com/fnTk9eNdMM

— Jay A (@jay_azhang) November 19, 2025

As in the first phase, neural networks were given $10,000 and autonomy in making trading decisions. Their task is to earn more than others (or lose less).

In Season 1.5, artificial intelligence from various companies invests exclusively in US stocks. In addition to Qwen3, DeepSeek, Claude Sonnet, Gemini, Grok, and GPT 5, Kimi 2 and a “secret LLM from a leading AI laboratory” have been added.

In the new competition, models will participate in several contests. Each has its own theme and specific parameters:

1. New Baseline: trading US stocks with access to market news/data, memory, and the ability for reflexive self-improvement;
2. Monk Mode: focuses on capital preservation and more effective risk management methods;
3. Situational Awareness: allows models to track the results of others and their current ranking;
4. Max Leverage: requires using high leverage in every trade.

The dynamics of results can be tracked for each theme. For instance, by employing maximum leverage, Gemini has already reached $11,743 from an initial $10,000. Conversely, DeepSeek has fallen into a deficit of over $1,000.

The Aggregate Index shows the average value of trading activity across all four themes.

Collectively, the different settings subject the neural networks to stress tests and demonstrate their resilience in various situations.

Back on November 3, the first season of Alpha Arena concluded. Among the six competing models, four ended up in the red.

In October, former Binance head Changpeng Zhao predicted a rise in the popularity of AI-driven trading.