PerspectiveGap Leaderboard
PerspectiveGap evaluates whether models can compose role-specific prompts for multi-agent orchestration without over-sharing context.
- Paper: arXiv:2606.08878
- Dataset: sun1245/PerspectiveGap
- Code and scorers: GitHub
This leaderboard summarizes the released paper sweep: 27 models × 220 rendered evaluations × 2 tasks.
Current top model: openai/gpt-5.5 with 62.0% combined pass rate.
This table contains all 27 models from the released paper sweep. The company tab is a 10-company summary, and the topology heatmap intentionally shows the top 12 models for readability.
Company