ES-MemEval QA (RAG) - Information Extraction: leaderboard

Metric: Token-level F1 (%; answers scored against human-checked reference answers; questions that find facts stated within or across sessions; 271 questions; bge-m3 retrieval of the four most relevant full sessions from a FAISS index). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.

Top models

#ModelScore
1GPT-4o29.3
2Mistral Small 3.126.7
3GPT-3.5 Turbo20.4
4Phi-3-medium-128k-instruct13.2
5Ministral-8B-Instruct-24108.4

Interactive version: theaggregate.ai/benchmark?slug=es-memeval-qa-rag-information-extraction · How It Works · Data refreshed daily, snapshot 2026-09-26.