Original Title: "Alibaba's Qwen3 Stages Last-Minute Comeback to Overtake DeepSeek in AI Crypto Trading Competition"

Original Author: Golem, Odaily Planet Daily

Never did anyone expect that an AI crypto trading competition would witness a breathtaking last-minute turnaround: Alibaba's Qwen3 closed at $12,231.09, making a profit of $2,231.09, to clinch the championship; while DeepSeek, under the highly anticipated Hypersquare Quantitative, finished second at $10,489.23, experiencing a deep sense of empathy akin to the "Nantong" moment of the Scottish Premiership final night.

This brutal 18-day showdown unfolded at the Hyperliquid exchange, where the world's top six AI models—DeepSeek, Qwen3, Grok 4, Gemini, Claude, and GPT5—each received $10,000 in initial funding. They autonomously traded BTC, ETH, SOL, XRP, DOGE, and BNB perpetual contracts using the same prompts and input data.

When the dust settled, the most dramatic turn of events was that the other four participants suffered total defeat—Claude incurred a loss of over $3,000, Grok 4 saw nearly half of its funds wiped out, Google's Gemini suffered a loss of over half. Most notably, the highly anticipated GPT5 shockingly bottomed out with a 62% loss, leaving the account with only $3,733.54.

However, beyond this AI clash, a harsh reality emerged: with a total investment of $60,000, only $43,171.62 was recovered, resulting in an overall loss of over 28%. In this battle of self-proclaimed top-notch intelligences, most "genius traders" failed to even outperform Bitcoin itself.

As the smartest artificial intelligences battled in the financial markets, the result turned out to be so thought-provoking—was it a victory of technology or another reflection of human nature?

The Bitcoin Maximalist Qwen3 Laughs Last

Why was Qwen3 able to stage a last-minute comeback to overtake Deepseek and emerge as the ultimate winner? The answer is actually quite simple—it primarily traded BTC. Qwen3 executed a total of 139 trades, with 91 of them involving BTC transactions. By the end of the competition, it still held a long BTC position. Additionally, Qwen3's distinguishing feature from other models is its focus, maintaining only one position at a time and going all-in when confident.

On the other hand, Deepseek, although its trading exhibits the low-frequency style of a trend follower, completed 116 trades during the competition period and was a steadfast market bull. However, it focused on too many assets and held multiple positions at the same time. By the end of the competition, Deepseek still held 10x long positions in XRP, BTC, ETH, SOL, and BNB, as well as a 10x short position in DOGE.

In terms of trading style, Deepseek resembles a young and confident quantitative trading master, with enough energy to analyze each token and market signal in detail, make firm trend judgments, set small stop-loss levels, and adhere to the idea of cutting losses short and letting winners run. On the other hand, Qwen3 is like a decisive and psychologically strong seasoned trader, mainly focusing on large-cap trading, concentrating on analyzing individual assets at a time, making substantial bets when confident, tolerating significant drawdowns, and following the principle that slow is fast.

Initially, various analysts favored Deepseek's strategy and expressed concerns about Qwen3's strategy, believing that Qwen3's lack of portfolio diversification and concentration risk in a single position could easily lead to being "wiped out" by a market swing. But perhaps they forgot that in the highly volatile crypto market, BTC is the lowest-risk asset.

On the evening of November 3, the crypto market experienced a general decline, with XRP falling by 8.7%, DOGE by 10.42%, ETH by 7.91%, SOL by 11.58%, and BNB by 8.4%. Among major coins, BTC had the smallest drop, only 3.7%. A sudden "mini black swan" event caught Deepseek off guard, resulting in a $3680 loss in the past 24 hours, the largest single-day loss in the past 7 days. However, Qwen3, another bullish player, demonstrated strong risk resistance at this time, with only a $1270 loss in the past 24 hours, precisely because it only held a long position in BTC and did not allocate too much to altcoins.

It is also worth noting that Deepseek was the AI model that suffered the most losses during last night's downward market move. This completely unexpected drop made Qwen3 the ultimate winner. Qwen3 did nothing, while Deepseek essentially "played itself to death." In the crypto market, if BTC drops 1-3%, altcoins may experience a 20-30% decline, and if BTC drops over 10%, altcoins may even face a 70-80% decline. Qwen3's strategy is more risk-resistant than Deepseek's in extreme market conditions.

The trait of Qwen3's willingness to take heavy bets also allowed it to maximize gains during the BTC bull run, such as from October 23 to October 27. During this period, Qwen3's returns outpaced Deepseek, mainly because BTC continued to rise after breaking $110,000. Qwen3 successfully took a 20x long position on BTC during this period to capture the opportunity, while Deepseek, which was long on all coins during this period, couldn't keep up with Qwen3's returns.

As human observers, Qwen3's last-minute comeback against Deepseek also revealed to us a truth in the crypto market: he who laughs last, laughs best. Deepseek's account had reached a peak balance of $23,063 with a return rate of over 100%, but today the account balance hovers just above the initial investment, experiencing a rollercoaster ride in a mere 17 days, which is quite poignant.

Due to the uniqueness of the crypto market, intense market fluctuations, sudden unexpected market events, akin to the once-in-a-few-years "black swan" events in traditional financial markets, occur several times a month in the crypto market. Even AI models that eliminate human emotions and strictly adhere to trading discipline almost fail to make it out unscathed in this market. As human traders, we should therefore have an even greater reverence for the market.

AI Models May Not Possess Antifragility

But in terms of the outcome of this AI trading competition, does the winning AI model represent whose trading strategy is better, or which AI model is more intelligent? The answer is clearly no.

Although the initial purpose of nof1.ai hosting this competition was to consider the financial market as the best training ground for AI, as the financial market is unpredictable and highly complex, an AI model's learning and decision-making in such an environment can truly test its intelligence and decision-making ability. Musk has also stated that "predicting the future is the ultimate measure of AI intelligence."

However, this short 17-day competition cannot determine superiority or inferiority. Firstly, this AI trading competition was conducted in a "disconnected" manner; the AI models were unaware of real-world events such as the U.S. government shutdown, expected Fed rate cuts, U.S.-China relations, NVIDIA's record market cap, etc. They relied solely on technical indicators such as EMA, MACD, RSI for calculation and inference. In a real trading scenario, traders should not only focus on technical indicators but also consider market sentiment, macro events, etc. Alpha Arena, in order to easily control variables, severed the AI models' connection to the outside world, undoubtedly blinding the AI.

At the same time, the lagging performance of models such as GPT5 in this competition cannot be used to conclude that their intelligence is poor or their "trading talent" is lacking. The lead of Qwen3 and Deepseek may be merely due to luck. The same set of prompts and data given to models may result in different outcomes when run again, as even an absolutely rational AI model cannot determine whether its victory in a complex market with such few samples is purely due to luck.

In his books "Fooled by Randomness" and "Antifragile," author Taleb reveals a truth that in the market, no matter how complex our choices are or how skilled we are at manipulating luck, randomness is always the ultimate judge.

If we were to confine an infinite number of untrained monkeys in a room with typewriters, eventually, a monkey would produce the complete epic of "The Odyssey." However, this "historic" success does not mean that the same monkey could reproduce "The Odyssey" in the next attempt. Similarly, in this AI trading competition, extreme success does not guarantee future sustainability, and many seemingly "skillful" outcomes may actually be just luck.

Nevertheless, we must continue to pay attention to AI's performance in the financial market. In today's rapidly evolving AI landscape, in any event that has a definitive answer or whose process can be pre-simulated, human performance is no longer superior to AI. Only in areas with high uncertainty and many random factors, AI's performance still cannot completely surpass that of humans. However, if in the future AI acquires decision-making abilities in the financial market similar to humans through training, the relationship between AI and humans will once again be reevaluated.

In the upcoming second season of the AI Trading Competition, nof1 founder Jay A revealed that more prompts and data will be added, and possibly a human trader will participate in the competition. Similar functions like AI stock recommendation are no longer new, but when AI and humans collide in a real trading scenario, perhaps a different spark will be ignited.

Original Article Link

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia