FULL STORY
Gemini Argon 4 Takes Second on MazeBench
Gemini Argon 4 scored 12% on MazeBench, ranking second behind GPT-6 Astra. The benchmark author followed up with heatmaps comparing exploration styles across top agents.
2026-10-15 ~ 2026-10-15 · 2 episodes · 5 posts
Episode 1 · Gemini Argon 4 scores 12% on MazeBench, second only to GPT-6 Astra (2026-10-15, 3 posts)
Gemini's Argon 4 scored 12% on MazeBench, becoming the fourth agent with non-trivial performance and ranking second behind GPT-6 Astra's 16%. Notably, it solved in one pass sokoban puzzles that GPT-6 Astra needed multiple attempts for, yet sometimes walked away when close to a solution.
- MazeBench: Gemini Argon 4 scores 12%, second only to GPT-6 Astra's 16% — patience_cave · 2026-10-15
- MazeBench: Argon one-shots puzzles that stumped GPT-6 Astra, but walks away near the finish — patience_cave · 2026-10-15
- Gemini team's new 3D puzzle agent becomes only the 4th with a non-trivial MazeBench run — patience_cave · 2026-10-15
Episode 2 · MazeBench heatmaps: Argon explores more thoroughly than GPT-6 Astra (2026-10-15, 2 posts)
MazeBench's author shared exploration heatmaps of Opus 5.5, Gemini 4 Argon and GPT-6 Astra, finding Argon explores more thoroughly and completely, while GPT-6 Astra covers similar rooms but in a more scattered way.
- MazeBench heatmaps: Argon's maze coverage beats Opus 5.5, rivals GPT-6 Astra — patience_cave · 2026-10-15
- MazeBench: Argon explores more thoroughly while GPT-6 Astra is stronger but sporadic — patience_cave · 2026-10-15