HELM NarrativeQA: leaderboard
Reading comprehension on full stories and movie scripts requiring free-form answers, scored by F1 overlap. Part of the Stanford HELM evaluation suite.
Metric: F1 (%). Source: crfm.stanford.edu. Status: saturated. 91 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4o (2024-05-13) | 80.39 |
| 2 | Llama 3 70B | 79.82 |
| 3 | DeepSeek V3 | 79.63 |
| 4 | GPT-4o (2024-08-06) | 79.54 |
| 5 | Llama 3.3 70B Instruct | 79.13 |
| 6 | Nova Pro | 79.13 |
| 7 | Gemma 2 27B (IT) | 78.97 |
| 8 | Gemini 1.5 Flash (001) | 78.29 |
| 9 | Gemini 1.5 Pro (001) | 78.25 |
| 10 | Yi 34B (Base) | 78.22 |
| 11 | Mistral Large 2 (Jul) | 77.87 |
| 12 | Mixtral 8x22B | 77.87 |
| 13 | Llama 3.2 90B Vision Instruct | 77.69 |
| 14 | Palmyra-X-004 | 77.26 |
| 15 | Llama 3.1 70B Instruct | 77.24 |
Interactive version: theaggregate.ai/benchmark?slug=helm-narrativeqa · How It Works · Data refreshed daily, snapshot 2026-09-05.