Adversarial Natural Language Inference — tests whether models can correctly determine if a hypothesis is entailed, contradicted, or neutral given a premise, using adversarially constructed examples.
This definition has not completed editorial and provenance review. Verify its metric, version, source, and evaluation configuration upstream before using it for a model-selection decision.
Adversarial construction targets known model weaknesses in logical reasoning.
Very few model scores (2). Older benchmark — may be largely saturated for frontier models. Score is 0–1 ratio.
Higher is better. Accuracy ratio (0–1).
Very few model scores (2).
| # | Model | Score |
|---|---|---|
| 1 | 0.6 | |
| 2 | 0.6 | |
| 3 | 0.6 |