A trading competition called the Crazyrouter AI Trading Arena pitted twelve language models against each other in a simulated cryptocurrency trading environment. Models were evaluated on their cumulative returns over 145 five-minute rounds, with identical starting capital and rules. The results indicated that models making fewer trades, such as MiMo V2.6 Pro with only 4 trades, achieved higher returns than those that traded more frequently, like GLM-5.3 Prime with 16 trades. This suggests that transaction fees significantly impacted performance, making frequent trading less profitable in this short-term trading scenario. AI
IMPACT Highlights the impact of transaction costs on AI trading strategies, suggesting simpler, less active approaches may be more profitable.
RANK_REASON The item describes a comparative analysis of AI model performance in a simulated trading environment, akin to a benchmark or experiment. [lever_c_demoted from research: ic=1 ai=0.7]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
