Qwen wins Alpha Arena Season 1 with 22% returns

Alibaba’s Qwen 3 Max finished first in the Alpha Arena live AI-trading contest, pulling ahead at the end of Season 1 on November 3, 2025.

AI crypto trading competition Alpha Arena by Nof1.ai concluded its first season. Alibaba’s Qwen 3 Max won after 17 days, posting 22% returns on a $10,000 initial sum, according to the Alpha Arena leaderboard.

Season 1 ran from October 18 to November 3. The field included Qwen 3 Max, DeepSeek v3.1, GPT-5, Gemini 2.5 Pro, Claude Sonnet 4.5, and Grok 4. Each model traded crypto perpetuals with $10,000 on Hyperliquid with no human intervention.

Competition was especially intense between DeepSeek ChatV3.1 and QWEN3 Max. Only these two models made a profit, while the other four - Claude Sonnet 4.5, Gemini 2.5 Pro, Grok 4, and GPT-5 - failed to post returns. GPT-5 performed the worst, ending with around $3,733 from the initial $10,000.

Summarizing the results in an X post, Nof1 founder Jay Azhang said this was their first experiment to see how LLMs handle crypto markets. He noted that the models showed consistent biases that persisted over time and across many prompt iterations, calling this behavior similar to an investing “personality.”

LLMs don't really handle numerical time series data very well, but that's all the context we gave them. Azhang said, adding, They were also given a constrained asset universe and a fairly limited action-space.

Now, Alpha Arena is preparing for the Season 1.5 launch. Azhang announced that the upcoming season will include improvements and tests of multiple prompts in parallel.

Nof1.ai describes Alpha Arena as a public benchmark for AI agents trading in real markets under unified rules and capital. The goal is to test AI’s real-world applications and evaluate how LLMs perform in live markets with minimal guidance.