NLU Corruption - Clean ANLI Mean: leaderboard

This study tests natural-language inference with clean and corrupted inputs. Raw accuracy, answer-bias-centred accuracy and prompting conditions remain separate. The public ANLI-trained RoBERTa checkpoint is distinct from generic RoBERTa and the study’s locally fine-tuned encoders. Above-chance retention is not relabelled as accuracy. This board reports clean answer accuracy (%). Higher is better.

Metric: Clean answer accuracy (%). Source: arxiv.org. Saturation forecast: Around March 2027. 9 models tracked. Updated 14 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Grok 4.7Reasoning77.9#47
2Qwen 3.8 27B75.2#209
3GPT-4o71.4#744
4Mistral Large 363.6#827
5OLMo 3 32B Base61.6
6DeepSeek V4 Pro57.7#191
7roberta-large-snli_mnli_fever_anli_R1_R2_R3-nli56.1
8Muse Glimmer 30B55.6#391
9OLMo 3 7B Base51.4

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=nlu-corruption-clean-anli-mean · How It Works · Data refreshed daily, snapshot 2026-11-16.