Wall Street is monitoring Nof1, an AI trading lab whose Alpha Arena pits frontier models from OpenAI, Anthropic, Google and xAI in live trading competitions. Most models lost money, but the failure is seen as a signal. SUI Group and Karatage co-led a $15M funding round.
DWF Ventures says automation and AI agents now make up about 19% of on-chain activity. They outperform humans in narrow yield strategies, but in trading contests, top human participants still beat the best agents by more than five times.
Alpha Arena’s live crypto trading test shows Qwen and Deepseek ahead after eight days, while GPT-5 has lost more than $7,000. The experiment highlights sharp differences in AI trading behavior, portfolio management, and risk control.
Nof1’s Alpha Arena puts six AI models into live crypto perpetual futures trading on Hyperliquid. After eight days, Qwen and Deepseek are leading, while GPT-5 is the weakest performer, underscoring both the promise and limits of AI-driven trading.
After eight days of live crypto futures trading on Hyperliquid, Qwen and Deepseek lead Alpha Arena while Western models remain in the red. The experiment highlights both the promise and limits of AI-driven trading.
In Nof1’s Alpha Arena crypto trading contest, four of six major LLMs finished with losses, led by ChatGPT after losing 63% of its capital. Only DeepSeek and QWEN3 MAX ended profitable, while trading fees and over-trading emerged as key drags on performance.
A crypto trading contest among six major LLMs found that four finished below their $10,000 starting balance, with ChatGPT posting the worst result at a 63% loss. Only DeepSeek and QWEN3 MAX ended in profit, while overtrading and fees weighed heavily on performance.
An AI crypto trading contest found that four of six major LLMs ended in losses, with ChatGPT performing worst after losing 63% of its starting funds. Only QWEN3 MAX and DeepSeek posted profits, while trading fees and overtrading hurt overall results.
