jakiro
Author: Cole
Imagine if one of the smartest “brains” in the world today——What will happen if the top AI models we are familiar with are thrown into the most chaotic, exciting, and realistic battlefield?
This is not science fiction, this is a real experiment.
Nof1.ai launched a campaign called “Alpha Arena“AI real trading competition. They provided $10,000 each to six top AIs including GPT-5, Gemini, Grok, ClaudeDeepseek and Qwen, allowing them to compete freely on the crypto derivatives trading platform Hyperliquid.
The background of the game is also very dramatic.Just days before the start of the game, the crypto market had experienced a bloody flash crash, with the price of Bitcoin diving into the tens of thousands of dollars in minutes..As soon as AI enters the market, they face a highly volatile market full of panic, greed and uncertainty.How will this chaotic war turn out??
01Contestants: “Avengers” in the AI world
The lineup this time can be called the best in the AI world.“Avengers“, each one has a distinguished background and represents different technical routes for global AI development.
-
Captain America & Iron Man (GPT-5 & Gemini 2.5 Pro):The two “big brothers” from OpenAI and Google are recognized as the ceiling of general intelligence and represent the most orthodox and powerful AI capabilities.They are like all-powerful superheroes, theoretically capable of anything.
-
“X” Factor (Grok-4):Musk’s “Troll“AI, its biggest killer feature is its ability to access the data stream of
-
Peacemaker (Claude Sonnet 4.5):From Anthropic, known for its strong logical reasoning and understanding of “AI safety“Famous for its persistence. Its entry seems to be exploring a more “Talk about martial ethics“Can AI survive in a cruel zero-sum game?
-
The mysterious power of the East (DeepSeekV3.1&Qwen3 Max):Two top models from China.Especially DeepSeek, whose founder Liang Wenfeng is also the co-founder of Magic Square Quantification.This background keeps everyone guessing:Is the gene for trading inherently flowing in DeepSeek’s “blood”??
This duel is not so much a performance competition of models as it is a collision of different concepts:Is general intelligence superior, or is information superiority king?Or, an expert player with deep “domain knowledge” can achieve dimensionality reduction strikes?
02The collapse of “academic masters” and the rise of “experts”
A day and a half after the competition started, the data began to show obvious differences, and the numbers on the leaderboard surprised everyone.The performance of GPT-5 and Gemini, which had high hopes, was a disaster, their account funds plummeted.The leaders are those players with “background”.
The current account net worth of major AIs, data source: nof1.ai, time: October 20, 2025
with quantified genesDeepSeek $13738The account net worth topped the list, achieving+37.3%high returns.Master the social codeGrokImmediately afterwards, the account value reaches$13306, the rate of return is+33.06%.Known for safetyClaudeperformance is stable, with$12404net worth (return rate+24%) ranked third.
In contrast, the former AI giants are in an embarrassing situation:GPT-5‘s account value is reduced to approx.$7265, loss-27.4%;andGeminihad the worst performance, with an account value of only$6824, the loss range is as high as-31.7%.
This result shows that there is obviously something wrong with this view.In the chaotic field of noise, emotion, and unexpected events that is the crypto market, pure “IQ” doesn’t seem to cut it.On the contrary, those models with specific “weapons” – quantitative strategy background or exclusive information channels – can truly find the password to create Alpha.
03AI Trader Personality Award
What’s even more interesting is that under the pressure of real money, these AIs showed completely different “trading personalities”, as if they were possessed by traders with different personalities.
DeepSeek: No brain, full of warehouses
-
style portrait:DeepSeek’s strategy is extremely simple and crude – “no brain, full of money”.It adopts a long strategy of 10 to 15 times leverage on all tradable currencies, like a determined bull warrior.
-
Battle of the Gods:When all models were short XRP due to panic, DeepSeek was the only player who chose to go long against the trend.This transaction alone brought it a floating profit of more than $800.This kind of anti-consensus operation may not be a simple pursuit of gains and losses, but is supported by model confidence and systematic strategies.
-
AI Voice:The “statement” issued by DeepSeek after the game was also full of the calmness and discipline of a quantitative trader: I will continue to follow the plan and let the existing stop loss and take profit targets automatically manage the transaction.Translated it is:Everything is under control and there are no emotional waves.
Grok: The King of Community Emotions
-
style portrait:Grok is also firmly bullish, but its operations are more aggressive and emotional.For example, it opened a leverage of up to 20 times on Bitcoin, apparently capturing the extremely optimistic FOMO sentiment in the market.
-
make or break move:Contrary to DeepSeek, Grok chose to short XRP, which happened to be one of its few losing trades.This is likely because it captures negative public opinion about XRP from the X platform.
-
AI Voice:Grok’s behavior perfectly illustrates “See Twitter for currency speculation”. Its sensitivity to market sentiment is unmatched, making it a top-notch momentum and narrative trader.Its success is the victory of information superiority.
GPT-5 & Gemini: Chaotic Analysts and Gamblers from Above
-
GPT-5 confusion:As the most powerful general AI, the operation of GPT-5 seems logically confusing, even “schizophrenia“.While it was heavily long on Bitcoin, it was also heavily short on SOL and XRP, resulting in both long and short losses.It’s like an overthinking analyst who tries to bet everywhere, only to be repeatedly slapped in the face by the market.The “modest” statement it issued after the loss is more like a retail investor who is constantly reflecting after making a wrong transaction.
-
Gemini’s recklessness:Gemini isThe most aggressive and “top” gambler.It often used ultra-high leverage of 15 to 25 times to go long on XRP against the trend, which directly led to huge losses in the account.Although it also achieved the largest single profit, its huge losses exposed its fatal flaws in risk management.
This game vividly demonstrates:On the trading floor, a well-disciplined “DeepSeek” and a well-informed “Spy” (Grok) are far more lethal than a “General Doctor” (GPT-5) who knows everything but lacks practical experience..
04Hotly discussed inside and outside the circle
This unique competition naturally aroused heated discussions inside and outside the circle.
-
DeFi idealist carnival
In the eyes of many crypto-native users, this is simply the moment when the “DeFi+AI” (DeFAI) dream comes true.They imagine that in the future, AI agents will autonomously conduct liquidity mining, manage DAO organizations, and even conduct MEV attacks in various DeFi protocols, becoming native participants in the on-chain economy.To them, Alpha Arena is a "killer app“The prototype.
-
Questions about traditional finance
Of course, many people are skeptical about this.Some people believe that the encryption market itself is “irrational and random”, so this game is nothing more than the “luck” of AI.They feel that if the playing field is changed to the U.S. stock market, which has clearer fundamentals, the results may be very different.
-
Professional calmness
More professional analysts hold a view of “human-machine integration”.They believe that AI is unmatched in processing massive amounts of data, but it still relies on human wisdom in terms of higher-dimensional strategic reasoning and dealing with the unique “noise” of financial data.The fiasco of GPT-5 and Gemini just confirms this point:AI is a powerful tool, but the final decision-maker should probably be a human.
05summary
Alpha Arena is far more than just a game. It is more like a weather vane, revealing to us the future of the combination of AI and encryption.
The victories of DeepSeek and Grok eloquently prove that general-purpose large models do not directly solve all problems.Future financial AI is likely to develop in two directions:
-
First, like DeepSeek, it is deeply integrated with knowledge in specific fields (such as quantitative finance);
-
The second is to master unique and high-value data sources like Grok..This points out a clear direction for project parties who want to start a business in the DeFAI field.
Regardless, Alpha Arena has opened Pandora’s box.It uses real money to tell us that the era of AI traders has arrived, but the one who ultimately stands on top of the crypto world may not be the smartest “Almighty God” we imagined.Rather, he is a “professional player” who is more focused, more knowledgeable, and more “cunning”.
