Multi-Agent Step Race Benchmark: LLM Collaboration and Deception Under Pressure (github.com)

interesting results, thank you!