We just published a reproducibility study of a paper about how LLM-based agents negotiate between each. The pre-print is already available on arXiv. Interesting read of you are into benchmarks that evaluate LLM capabilities applied to negotiation.