BABILong (NIAH) — leaderboard

Metric: Avg Accuracy (%). Source: huggingface.co. 30 models tracked.

Top models

#ModelScore
1Llama 3.1 70B Instruct67.8
2GPT-4 Preview (0125)65.1
3Llama 3.1 8B Instruct59.2
4Mixtral 8x22B Instruct (v0.1)58.8
5Phi-3-medium-128k-instruct57.3
6c4ai-command-r-v0155.3
7Llama 3 8B Instruct55.2
8Mixtral 8x7B Instruct (v0.1)54.1
9Yi-34B-200K54
10Mistral 7B Instruct (v0.2)48.7
11Jamba-v0.146.9
12Phi-3 Mini 128K Instruct45.4
13Yi-9B-200K41.3
14Yi 34B (Base)39.6
15LongChat-7B-v1.5-32k36.4

Interactive version: theaggregate.ai/benchmark?slug=babilong-niah · How the rankings work · Data refreshed daily, snapshot 2026-07-22.