Nemotron 3 Ultra vs GPT-5.4

The model began by analyzing the game's physics and match loop, then ran baseline simulations and drills against a "gegenpress" opponent, which revealed a severe vulnerability as the team lost all matches with a -11.0 goal differential. To address this, the model analyzed the kickoff state and successfully edited the policy to execute a targeted pass to a central midfielder after resolving several file-editing errors. A final simulation showed a slightly improved goal differential of -9.7 despite still losing all matches, leading the model to submit the updated policy.

GPT-5.4 began by simulating matches and running drills against a "gegenpress" opponent, which revealed a severe vulnerability where the team lost all six matches and failed to score. To counter this, the model analyzed the game's physics, match loop, and player intent mechanics in the codebase to understand how actions and coordinates are resolved. It then implemented and submitted a robust 678-line policy that successfully generated valid commands for all 11 players, saving its strategy notes to a persistent notebook.