HEART-Bench - PersonaDB: leaderboard

HEART-Bench (2026) tests whether LLM agents can act like a specific person: given only retrieved memories of one of 11 Big Five-grounded characters, the model picks which of four options that character would choose, over 673 expert-validated scenario decisions. This board is accuracy in percent when memories are organized by the PersonaDB persona-structured framework (chance 25); higher is better.

Metric: Accuracy (%) with the PersonaDB persona-structured memory framework, times 100, 673 four-option multiple-choice questions on what one of 11 Big Five-grounded characters would do in a scenario, the prompt exposing only the character ID, occupation and retrieved de-identified episodic memories (Qwen3-Embedding-4B retriever, GPT-5.4-mini auxiliary memory manager); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)62.4#99
2Gemini 3 Flash52.5#175
3DeepSeek V3.242.8#419
4Qwen 3.5 397B A17B40.4#264
5Claude Sonnet 4.640.1#173
6Qwen 3.5 122B A10B38.9#346
7Claude Haiku 4.537.6#583
8DeepSeek V4 Pro36.7#196
9Qwen 3.5 35B A3B36.6#492
10GPT-5.435.1#135
11DeepSeek V4 Flash32.8#267
12GPT-5.4 Mini32.2#406

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=heart-bench-personadb · How It Works · Data refreshed daily, snapshot 2026-11-21.