BABILong: leaderboard
BABILong evaluates long-context language models on synthetic QA tasks over increasingly long contexts, reporting accuracy by context length.
Metric: Mean <=128k (%). Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.1 70B Instruct | 67.78 | #561 |
| 2 | GPT-4 Preview (0125) | 65.11 | #486 |
| 3 | Llama 3.1 8B Instruct | 59.22 | #1035 |
| 4 | Phi-3-medium-128k-instruct | 57.33 | #1013 |
| 5 | c4ai-command-r-v01 | 55.33 | #989 |
| 6 | Mixtral 8x22B Instruct (v0.1) | 52.22 | #756 |
| 7 | Yi-34B-200K | 48 | #1176 |
| 8 | Phi-3 Mini 128K Instruct | 45.44 | #1298 |
| 9 | Mixtral 8x7B Instruct (v0.1) | 42.11 | #1073 |
| 10 | Yi-9B-200K | 41.33 | #1204 |
| 11 | Mistral 7B Instruct (v0.2) | 37.89 | #1333 |
| 12 | Yarn-Mistral-7B-128k | 32.11 | #1398 |
| 13 | Yi 34B (Base) | 30.78 | #972 |
| 14 | Llama 3 8B Instruct | 30.67 | #1129 |
| 15 | Llama 2 7B 32K Instruct | 30.33 | #1508 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=babilong · How It Works · Data refreshed daily, snapshot 2026-10-20.