BABILong (NIAH): leaderboard
Metric: Avg Accuracy (%). Source: huggingface.co. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3.1 70B Instruct | 67.8 |
| 2 | GPT-4 Preview (0125) | 65.1 |
| 3 | Llama 3.1 8B Instruct | 59.2 |
| 4 | Mixtral 8x22B Instruct (v0.1) | 58.8 |
| 5 | Phi-3-medium-128k-instruct | 57.3 |
| 6 | c4ai-command-r-v01 | 55.3 |
| 7 | Llama 3 8B Instruct | 55.2 |
| 8 | Mixtral 8x7B Instruct (v0.1) | 54.1 |
| 9 | Yi-34B-200K | 54 |
| 10 | Mistral 7B Instruct (v0.2) | 48.7 |
| 11 | Phi-3 Mini 128K Instruct | 45.4 |
| 12 | Yi-9B-200K | 41.3 |
| 13 | Yi 34B (Base) | 39.6 |
| 14 | mamba-2.8B-hf | 33.2 |
| 15 | GPT-2 | 21 |
Interactive version: theaggregate.ai/benchmark?slug=babilong-niah · How It Works · Data refreshed daily, snapshot 2026-09-05.