BABILong (NIAH, v0, 8k) Leaderboard · The Aggregate

BABILong (NIAH, v0, 8k): leaderboard

Metric: QA1–5 Mean Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 29 models tracked.

Top models

#	Model	Score
1	GPT-4 Preview (0125)	71
2	Llama 3.1 70B Instruct	70
3	Llama 3.1 8B Instruct	62
4	Phi-3-medium-128k-instruct	60
5	c4ai-command-r-v01	59
6	Mixtral 8x22B Instruct (v0.1)	58
7	Yi-34B-200K	52
8	Mixtral 8x7B Instruct (v0.1)	50
9	Phi-3 Mini 128K Instruct	50
10	Mistral 7B Instruct (v0.2)	45
11	Yi-9B-200K	45
12	Llama 3 8B Instruct	44
13	LongChat-7B-v1.5-32k	42
14	Llama 2 7B 32K Instruct	40
15	Llama 2 7B 32K	39

Interactive version: theaggregate.ai/benchmark?slug=babilong-niah-v0-8k · How It Works · Data refreshed daily, snapshot 2026-10-09.