I gave Jev, Laya, finetuned ModernCE and Qwen3.5 the controls to Doom

Researchers tested Jev, Laya, finetuned ModernCE-base-nli, and finetuned Qwen3.5-4B models by giving them control of Doom. The finetuned Qwen3.5-4B model achieved the highest mean kills at 3.63, while Laya English had the longest mean survival time at 11.89 seconds. Call latency varied, with finetuned ModernCE-base-nli showing the lowest p50 at 7.6 ms. These models received no Doom-specific training for these evaluations.

This report uniquely details a comparative performance evaluation of four distinct AI models controlling Doom, unlike typical benchmarks focusing on single-model capabilities.