ConvRe - Re2Text Hard: leaderboard
The "Re2Text Hard" subtask of ConvRe. ConvRe evaluates language models' understanding of converse relations through Re2Text and Text2Re tasks in easy and hard settings.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | text-davinci-003 | 55.8 | #1146 |
| 2 | GPT-4 (0314) | 16.4 | #574 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=convre-re2text-hard · How It Works · Data refreshed daily, snapshot 2026-10-20.