← back to paper

Review history

arxiv: 2608.02046 · 2 revisions

CompanionBench: A Theory-Anchored, Real-World-Grounded Benchmark for AI Emotional Companionship

1. 2026-08-07 CONDITIONAL MODERATE v1.4.0-alltime-deepseek-medium novelty 8.0
   169532 ms 37955 in 17305 out 2026-08-07T00:09:49.754206+00:00
2. 2026-08-04 CONDITIONAL HIGH v1.3.0-daily-deepseek novelty 8.0
   113737 ms 37762 in 13878 out 2026-08-04T16:06:57.380078+00:00