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: huggingface.co. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 91 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-05-13) | 80.39 | #325 |
| 2 | Llama 3 70B | 79.82 | #631 |
| 3 | DeepSeek V3 | 79.63 | #321 |
| 4 | GPT-4o (2024-08-06) | 79.54 | #336 |
| 5 | Llama 3.3 70B Instruct | 79.13 | #534 |
| 6 | Nova Pro | 79.13 | #569 |
| 7 | Gemma 2 27B (IT) | 78.97 | #645 |
| 8 | Gemini 2.0 Flash (Preview) | 78.3 | #370 |
| 9 | Gemini 1.5 Flash (001) | 78.29 | #577 |
| 10 | Gemini 1.5 Pro (001) | 78.25 | #444 |
| 11 | Yi 34B (Base) | 78.22 | #972 |
| 12 | Mistral Large 2 (Jul) | 77.87 | #484 |
| 13 | Mixtral 8x22B | 77.87 | #766 |
| 14 | Llama 3.2 90B Vision Instruct | 77.69 | #622 |
| 15 | Llama 3.1 70B Instruct | 77.24 | #561 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-narrativeqa · How It Works · Data refreshed daily, snapshot 2026-10-20.