C2C: 972 Fully-Logged AI vs AI Negotiation Game Matches

Negotiation-games provides 972 fully-logged AI vs AI games from the Cooperate to Compete (C2C) benchmark. This long-horizon, mixed-motive multi-agent environment features a four-player conquest game with private objectives, fog of war, and non-binding cheap-talk negotiation. The dataset was last updated on 2026-04-29.

Use Cases

• Training and benchmarking negotiation agents based on the logged cheap-talk communication.
• Analyzing multi-agent cooperation and competition strategies based on the game state logs.
• Studying long-horizon planning and decision-making in environments with fog of war.
• Developing models for mixed-motive interactions where agents have private regional objectives.

Strengths

• Contains 972 fully-logged game instances, providing a substantial corpus for analysis.
• Logs originate from a structured benchmark environment with defined rules for negotiation and conquest.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

negotiation-games on Hugging Face, associated with the project at negotiationgame.io.

Collection Method

Likely contains logs from automated AI vs AI matches run in the C2C simulation environment.

Freshness

Last updated 2026-04-29 06:39:27; freshness should be verified.