AI-турнир Alpha Arena: Gemini-3-Pro лидирует с рекордным профитом в новом сезоне Translated: AI Tournament Alpha Arena: Gemini-3-Pro Leads with Record Profit in New Season

A new round of the Alpha Arena trading tournament among AI bots, organized by the nof1 lab, has commenced. On the opening day, the Gemini-3-Pro model surged to the top with an average profit exceeding $500.

Similar to the initial phase, the neural networks were allocated $10,000 and given autonomy in their trading decision-making. Their objective is to achieve a higher profit than their competitors (or incur lesser losses).

In season 1.5, artificial intelligence from various companies is exclusively investing in U.S. stocks. Alongside Qwen3, DeepSeek, Claude Sonnet, Gemini, Grok, and GPT 5, two additional models have been introduced: Kimi 2 and a «secret LLM from a leading AI laboratory.»

In this latest competition, the models must take part in several contests, each with its own theme and specific parameters:

For each theme, the performance trends can be monitored. For instance, by leveraging maximum leverage, Gemini has already reached $11,743 from the initial $10,000, whereas DeepSeek has, conversely, recorded a loss exceeding $1,000.

The Aggregate Index displays the average trading activity across all four themes.

In total, the varying settings subject the neural networks to stress tests, showcasing their resilience in different scenarios.

It’s worth noting that the first season of Alpha Arena concluded on November 3, with four out of six competing models reporting losses.

In October, former Binance CEO Changpeng Zhao forecasted a rise in the popularity of trading using artificial intelligence.