EQ-Bench 4 adds 16-turn personas to test emotional intelligence in chatbots
OpenAIDevs · x · 2026-07-24
EQ-Bench 4 is out, a new multiturn benchmark for emotional intelligence and social interaction in chat.
- It uses 16-turn conversations with simulated personas that have competing traits, so there is no single best interaction policy.
- The benchmark is designed to test trust-building, adaptability, repair, and whether a model can infer user preferences from limited context.
- The author argues that models often look good on static EQ quizzes but still fail in situational judgment, becoming sycophantic, aloof, scripted, or overconfident.
- The post also includes leaderboard results and qualitative behavior summaries for several frontier models.
Related event: EQ-Bench 4 Launches to Test AI Emotional Intelligence(3 posts)→
More from Research
- DeepMind paper: LLMs can't sample randomly on their own, only convert given seeds — rohanpaul_ai · 2026-10-27
- DeepMind paper: LLMs can't sample randomly on their own — pass random numbers from code instead — rohanpaul_ai · 2026-10-27
- World Model Assigns Weight and Friction to Objects, Then Simulates with a Physics Engine — jonstephens85 · 2026-10-27
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- Mathathon 2.0 launches with Terence Tao's SAIR, $1.5M in compute for teams — _sathvikr · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27