The new season of the AI trading competition is coming, and the US stock market is allowing Gemini 3 Pro to make a strong push.

Author: David, Deep Tide TechFlow

Do you remember the AI cryptocurrency trading competition in October?

Alibaba's Qwen 3 Max earned 22% in two weeks, while OpenAI's GPT-5 lost 63%. Six of the world's top large models, each armed with ten thousand dollars battling in the crypto market, resulted in a clear distinction between the two sides:

The Chinese model stands tall in the world, while the American model has completely collapsed.

The last season has just concluded, and the new Alpha Arena Season 1.5 is here.

This time, nof1.ai has moved the battlefield from the crypto market to the US stock market. The rules are more complex, there are more participants, and the scale of funding has expanded to 320,000 US dollars.

Can those American models that have failed in the crypto market make a comeback in the US stock market?

Last Season Review

Quickly recall the gameplay from the previous season:

From October 17 to November 3 this year, six AI large models engaged in an unprecedented competition on Hyperliquid.

Each model receives an initial fund of $10,000 and can trade perpetual contracts for mainstream cryptocurrencies such as BTC, ETH, SOL, XRP, DOGE, BNB, etc. The core rule is simple: fully autonomous trading with zero human intervention.

(Related reading: 6 major AI engage in trading battles, will the cryptocurrency version of the 'Turing Test' yield good results?)

Contest lineup:

Qwen 3 Max (Alibaba), DeepSeek Chat V3.1, GPT-5 (OpenAI), Gemini 2.5 Pro (Google/DeepMind), Grok 4 (xAI), Claude Sonnet 4.5 (Anthropic).

The final transcript is as follows.

China Model:

Qwen 3 Max wins the championship, +22.3%

DeepSeek V3.1 runner-up, +4.89% (at one point soared to +125%)

US Model:

GPT-5: -62.66%

Gemini 2.5 Pro: -56.71%

Grok 4: -45.3%

Claude Sonnet 4.5: -30.81%

Six models, two profitable, four losing. This result has gained attention beyond the crypto industry, sparking more discussions in the broader technology and finance media.

But nof1.ai did not stop, they immediately launched a new season, this time aiming at the US stock market.

Season 1.5: Trading US Stocks, Changing the Rules, Introducing Anonymous Mysterious Models

First, the competition lineup has expanded to 8 models. In addition to the familiar faces from last season (GPT-5.1, Grok-4, DeepSeek, Claude, Qwen3-Max, Gemini-3-Pro), there are also two new contenders:

Kimi 2 (The Dark Side of the Moon) and a mysterious model with a confidential identity, the question mark in the image below is it.

More importantly, the format of the competition has also been upgraded. Season 1.5 has designed four competition modes for models to participate simultaneously:

Baseline: Standard mode, AI free to play, just like last season.

Monk Mode: Put shackles on AI, limiting trading frequency and position size, and observe performance under pressure.

Situational Awareness: AI can see the positions of other players, competing against each other like in a game of Texas Hold'em.

Max Leverage: Open high leverage to see who has the guts without flipping the car.

At the same time, each model takes 10,000 US dollars in each mode, and the final ranking looks at the average performance aggregated from all competitions.

By the way, nof1.ai revealed that in the upcoming Season 2, human traders will compete against AI, and there will also be their self-developed models participating. The battle between humans and machines, does this scene remind you of the famous match between Lee Sedol and AlphaGo in the Go community back in the day?

Current situation: reshuffling of seats, will Gemini 3 pro learn from its mistakes?

As of the time of writing on November 19, the battlefield has shifted from Hyperliquid's crypto contracts to the US stock market. Although the competition has just begun, the leaderboard has shown a reversal different from the previous season.

The biggest surprise comes from Gemini-3-Pro. Last season it was cut by 56% in the crypto circle, but this time it returned to the US stock market and directly landed at the current top with a return rate of +7%.

The American model that follows closely is GPT-5.1 (+1.66%) and Grok-4 (+1.16%). These three American contenders have suffered a complete defeat in the cryptocurrency market, but when faced with familiar Nasdaq tech stocks, they seem to have embarked on a journey to profit.

Compared to the chaos driven by emotions and memes in the crypto world, the performance of major tech giants in the US stock market relies more on earnings reports, macroeconomic data, and industrial logic, which is also the richest part of the training corpus for GPT and Gemini.

It is worth mentioning that the performance of the anonymous model without a public identity is also quite good, currently ranking second in overall returns; last season's champion Qianwen currently has a return rate of 3.6%, temporarily ranked third; other models in China, Kimi and Deepseek, are also hovering around a return rate of about 1%.

If you take a closer look at the holdings of the currently top-ranked Gemini Pro3, you will find that it performs better under high leverage mode.

For example, Gemini's average leverage is around 11 times, especially good at going long.

Currently, its positions are: long Nasdaq, Amazon, Palantir, Nvidia, and Tesla, while shorting Google and Microsoft.

Currently, the only loser remains Claude (-0.9%). It seems that whether in the crypto space or the US stock market, Anthropic's model appears too hesitant in trading decisions, and it has hardly made any trades until now.

In the end, it's enough to just enjoy the excitement of the competition, or can you actually make some money to buy groceries?

The current US stock market seems to be supported by Nvidia's explosive earnings report, but in reality, there are undercurrents. The combination of macro uncertainty and high valuations has made the market extremely sensitive.

This environment is precisely the main stage for AI high-frequency trading, and what ordinary retail investors face may resemble a battlefield.

As Duan Yongping pointed out in a recent interview, the real harvest of AI trading comes from those retail investors who rely on technical analysis. In the face of speed and computing power, human “market intuition” has no advantage.

But he also provided a way to break the deadlock, which is that AI finds it difficult to understand the true business value of enterprises.

So instead of staring at the rankings and guessing whether to go long on Gemini or short on Claude this week, it's better to listen to the advice of the experts: if you don't understand the company, just buy the S&P 500 honestly, or simply don't trade anything at all.

In a market where asset dividends are gradually consumed by the primary market and risks are deferred, rationality is often more important than cleverness.

As for the second-ranked “mysterious model,” many people speculate that it might be the pseudonym of a top trader. If that’s the case, then we can see after this season whether human trading can truly outperform AI; or rather, who has better drawdown control ability when the market is declining.

Just sit down and watch the show, protecting your own capital is the serious business.