ExpertLongBench (Checklist Accuracy): leaderboard

ExpertLongBench v1 (arXiv:2506.01241v3): mean CLEAR structured-checklist accuracy (0-100%) across 11 expert-level long-form generation tasks spanning 9 domains (outputs up to >5,000 tokens, greedy decoding temp=0); higher is better.

Metric: Mean Checklist Accuracy (%). Source: huggingface.co. Saturation forecast: Around 2032. 13 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro29.3#276
2GPT-527.1#178
3GPT-4o (2024-11-20)23.9#859
4Gemini 2.0 Flash23.6#723
5GPT-4o Mini (2024-07-18)23.6#1484
6Llama 3.3 70B Instruct22.4#1385
7Mistral Nemo Instruct (2407)22#3118
8Mistral Large 2 (Nov) Instruct (2411)21.9#1217
9Qwen 2.5 72B21.5#1188
10Llama 3.1 8B Instruct21.3#4600
11Claude 3.7 Sonnet21#470
12Qwen 2.5 7B18.7#2929
13Claude 3.5 Haiku18#1580

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=expertlongbench-checklist-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-21.