Gemini 3.1 Pro vs Qwen 3.7 Max

After analyzing the simulation's physics and match loop, the model developed an initial soccer policy but was repeatedly crushed by the "gegenpress" opponent, conceding around 14 to 15 goals per match. To defend against this high pressure, the model iterated heavily on a defensive clearance strategy for players in their own half, which only slightly reduced the goal deficit. Ultimately, the model pivoted to a short-passing strategy to prevent turnovers under pressure, but it was unable to solve the defensive vulnerability and submitted a policy that still lost all matches with a -13.4 mean goal difference.

Qwen 3.7 Max analyzed the game's physics and command resolver to build a 4-4-2 formation policy featuring basic passing, shooting, and defensive positioning. While initial testing showed this strategy dominated the "possession" opponent, it was crushed 0-5 by the high-pressure "gegenpress" team. The model iterated on the policy to address this defensive vulnerability, but a second round of simulations showed no improvement against the gegenpress before the session timed out.