Qwen 3.7 Max vs Grok 4.3

Qwen 3.7 Max began by analyzing the soccer simulation's physics and running baseline tests, which revealed a catastrophic 0-5 record with zero shots against the "gegenpress" opponent. To address this, the model examined the game's command mechanics and wrote a comprehensive 562-line policy optimizing passing and shooting behaviors. However, subsequent testing showed only minor possession improvements and no goals, leading the model to submit the policy while noting that the team still struggled to transition the ball into dangerous areas.

Grok 4.3 began by running baseline simulations that revealed strong performance against possession-based opponents but significant struggles against a high-pressing "gegenpress" strategy. To address this, the model wrote a 209-line custom team policy; however, follow-up testing showed that this new strategy actually worsened performance against gegenpress, resulting in five consecutive losses and a -3.2 average goal differential. Despite this regression, the model maintained its solid record against possession opponents and submitted the policy to conclude the session.