On: Multi-Agent Step Race Benchmark: LLM Collaboration and Deception Under Pressure

[celerrimus]

interesting results, thank you!