China's DeepSeek and Qwen AI Beat US Rivals In Crypto Trading Contest

hackingbear shares a report from Crypto News: Two Chinese artificial intelligence (AI) models, DeepSeek V3.1 and Alibaba's Qwen3-Max, have taken a commanding lead over their US counterparts in a live real-world real-money cryptocurrency trading competition, posting triple-digit gains in less than two weeks. According to Alpha Arena, a real-market trading challenge launched by US research firm Nof1, DeepSeek's Chat V3.1 turned an initial $10,000 into $22,900 by Monday, a 126% increase since trading began on October 18, while Qwen 3 Max followed closely with a 108% return.

In stark contrast, US models lagged far behind. OpenAI's GPT-5 posted the worst performance, losing nearly 60% of its portfolio, while Google DeepMind's Gemini 2.5 Pro showed a similar 57% decline. xAI's Grok 4 and Anthropic's Claude 4.5 Sonnet fared slightly better, returning 14% and 23% respectively. "Our goal with Alpha Arena is to make benchmarks more like the real world -- and markets are perfect for this," Nof1 said on its website.

Well of course

[RitchCraft]

We all know the Chinese would never cheat.

Re:

[sabbede]

Or we could just say the true thing - that the CCP cannot be trusted. I wouldn't be shocked if they manipulated the markets in order to guarantee an outcome. I recall there being quite a lot of crypto held in China, though that may no longer be true.

Sounds like a very poorly disguised advertisement

[ffkom]

... for scammy crypto-things sold to gullible people. But even if that "contest" had any other background, the results would hardly be reproducible at any future point in time. You could just as well judge LLMs by their ability to predict dice rolls.

Re:

[Nrrqshrr]

Exactly. The stock market is so disconnected from reality that it's basically a casino. The cryptocurrencies market is even worse.
I would rather see them pit these AIs in something actually measurable, but then we would all realize that these are just tools and there's no inherent understanding behind their decisions.

Re:

[ffkom]

That LLM poker tournament [pokerbattle.ai] is probably more entertaining, and unlike the crypto scam does not pretend to be about anything but gambling.

Re:

[retchdog]

ackshually... casinos are much more tightly regulated.

Still waiting

[Waffle Iron]

I'll know that AI has achieved a major milestone when one of them eventually says:

"Crypto is a strange game. The only winning move is not to play."

Re:

[Waffle Iron]

Are they smart enough to know to pull those gains out before the tulip market crashes?

Re:

[ArchieBunker]

I’ve been hearing slashdot talk about tulips for a solid decade now. No sign of collapse in sight.

Re:

[Waffle Iron]

It was demonstrated long ago that most all trading markets are chaotic. By definition, you won't see a sign of collapse in sight.

Unlike useful markets, (but similar to 17th century tulips), crypto has no intrinsic value. Therefore, there is no bottom to a potential crash.

Re:

[DamnOregonian]

Correction:

There is potentially no bottom to a crash.
In practice, however, there is. And it's been demonstrated.

Of course, the next one could be different.

Re:

[DamnOregonian]

A crash could be decades out, or not at all.
The market works on investor confidence, and there are a lot of believers.

As for wise enough to pull money out, wellllll...
The Chinese LLMs are ahead, so they've certainly proven to be smart enough to pull their money out to profit in the short term, while one set of US LLMs seems to be performing just about as wall as HODL, and 2 of them seem to be pulling out low and buying high, or something nonsensical.

Frankly, the 2 Chinese bots continued performance in

Random Walk

[quantaman]

Ignoring the whole fact that we don't even know how they set up these bots to trade, it looks pretty damn random.

Re:

[DamnOregonian]

It's all above board. Pretty standard LLM harness. Go read about it. The runners of the competition aren't invested in an outcome, it's just a game.

While the game operates fairly, I think that might actually be the wrong way to go about this.
Every LLM I've ever worked with in a harness (agentically) has had quirks for how you need to "talk" to it to get it to perform best.
It's very possible that there is an unintentional bias in harness performance for the Chinese LLMs.

Re:

[DamnOregonian]

Also, no. Not random at all.

You have a couple bots with significant gains, a couple of bots that have basically followed the market, and a couple of bots that have vastly underperformed the market.

The middle group are performing at basically the level of a random coin flip.
The top group are outperforming a coin flip, and the bottom group are underperforming a coin flip.
What's important is that they're consistently doing so, over time.

The existence of the upper and lower groups indicates that the und

Fire Idea

[dohzer]

When cryptocurrency isn't wasting enough energy, you can always throw AI at it.

Re:

[DamnOregonian]

Ya, that part is pretty funny.
This might be the most wasteful endeavor ever undertaken, beyond a Republican energy plan.

need to know risk

[buddyglass]

Kinda need to know what level of risk the Chinese AIs took on. Getting big gains by making risky bets and getting lucky = meh.

Re:

[DamnOregonian]

Luck is random. The results are clearly not random.

Rather, if risk plays a factor, it means the market on average favors risk takers.
i.e., they will come out ahead.

Some extremely expensive programs playing dice

[doragasu]

You would be surprised some get better guesses than other.

Re:

[DamnOregonian]

If they were playing dice, you'd be right. However, they're not.