HEART-Bench: 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 averaged over three memory frameworks, Naive-RAG, Mem0 and PersonaDB (chance 25); higher is better.

Metric: Accuracy (%) averaged over the three memory frameworks (Naive-RAG, Mem0 and PersonaDB), 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)62.7#99
2Gemini 3 Flash54.3#175
3DeepSeek V3.242.2#419
4Qwen 3.5 397B A17B40.4#264
5Claude Sonnet 4.639.7#173
6Qwen 3.5 122B A10B39.1#346
7DeepSeek V4 Pro38.2#196
8Qwen 3.5 35B A3B36.9#492
9Claude Haiku 4.536.5#583
10GPT-5.436.3#135
11DeepSeek V4 Flash33.7#267
12GPT-5.4 Mini33.2#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 · How It Works · Data refreshed daily, snapshot 2026-11-21.