Episodic Memory — leaderboard
Tests how well LLMs encode, store, and recall episodic events across long narratives (200 chapters, ~100K tokens, 686 Q&A pairs). Measures simple recall and chronological awareness.
Metric: Simple Recall (%). Source: github.com. Status: saturation imminent. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 96.8 |
| 2 | Gemini 2.5 Flash | 96 |
| 3 | GPT-5 | 94.2 |
| 4 | GPT-5 Mini | 83 |
| 5 | Claude Sonnet 4 | 79 |
| 6 | Grok 4 Fast (Reasoning) | 72.6 |
| 7 | Gemini 2.0 Flash (Thinking) | 70.8 |
| 8 | GPT-4o | 67 |
| 9 | Grok 4 Fast (Non-reasoning) | 60.2 |
| 10 | DeepSeek V3 | 60 |
| 11 | Gemini 2.0 Flash | 59.6 |
| 12 | DeepSeek R1 | 57.2 |
| 13 | Llama 3.1 405B | 50.4 |
| 14 | GPT-4o Mini | 49.2 |
| 15 | Claude 3.5 Sonnet | 47 |
Interactive version: theaggregate.ai/benchmark?slug=episodic-memory · How the rankings work · Data refreshed daily, snapshot 2026-07-22.