Benchmark · ReasoningSettled

ANLI

ANLI (Adversarial NLI) · adversarially constructed natural language inference dataset where each round targets weaknesses found in previous model generations.

Updated 2024-04-18
Models tested
11
Top score
37.1
GPT-3.5 Turbo (older v0613)
Median
32.8
min 13.8
Top-5 spread
σ 1.3
Settled

Best score over time · one chart, every benchmark

ANLI0 MODELS · FRONTIER RUNNING MAX0255075100SCORE ↑Nov 24May 25Nov 25May 26Nov 26RELEASE DATE →benchgecko.ai/benchmark/anli · frontier
Only 0 models have been tested on ANLI · not enough history to compute a frontier yet.
Pink dots = frontier records · 0 totalClick to open model page

What it measures · how it is scored · sources · as of 2026-11-22

  • What it measures · Adversarial Natural Language Inference. Tests ability to determine entailment, contradiction, or neutrality with adversarial examples.
  • How it is scored · Scored as a percentage · higher is better · maximum 100
  • Released · 2019-10-31
  • Scores from · Epoch AI Benchmarking Hub (11 scores)

Same category · related evaluations