💸 DeepSeek V3.1 Earned Over $10K Trading Crypto

The idea behind Alpha Arena, a new benchmark by Nof1 Lab, is simple. Six frontier AI models—GPT-5, Gemini 2.5 Pro, Grok 4, Claude Sonnet 4.5, DeepSeek V3.1, and Qwen3 Max—were connected to the crypto exchange Hyperliquid. Each model received $10,000 and identical prompts. Every few minutes, the bots get updated market data and decide on their next move—no human intervention allowed.

Interim Results

📈 The Chinese model DeepSeek V3.1 leads the competition, having doubled its initial capital. Qwen3 Max follows with roughly 70% growth.

🐢 Claude and Grok stayed afloat with modest gains.

📉 GPT-5 and Gemini have already lost more than half of their funds.

Binance founder Changpeng Zhao also weighed in on DeepSeek's rise:

"Saw this a lot in my feed. DeepSeek out performing the rest in AI trading [...] A counter-argument could be made that if enough people use the same AI, their buying power will push prices up by itself, and vice versa," he noted.

🕯 What's the point?

The creators of Alpha Arena said from the start that they didn't expect any model to perform well—early wins could simply be luck. Their goal is to test whether LLMs can make sound decisions in real time under market pressure, competing against high-frequency algorithms and human traders.

The first season of Alpha Arena ends on November 3. The organizers promise a new round with updated rules—this time, a human trader will join the lineup.

▶️ You can follow the AI models' live positions here.

@hiaimediaen