Neo-Classic - Word Cloze (Random Distractors): leaderboard
Cloze: restore a masked two-character word from four tone-consistent candidates sampled without part-of-speech linkage, on a contemporary out-of-sample corpus of metrically verified classical Chinese poems by pseudonymous university poets, from Neo-Classic (Zhang et al., 2026). Accuracy in percent on a 100-question sample; higher is better.
Metric: Accuracy (%; restore a masked two-character word in a contemporary poem from four tone-consistent candidates sampled without part-of-speech linkage; accuracy on a 100-question sample of the task, as reported in the paper's full results table (accuracy@IF); temperature 0.1, top-p 0.9). Source: arxiv.org. Saturation forecast: Around 2029. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 Max | 69 | #401 |
| 2 | DeepSeek V3.2 | 65.4 | #403 |
| 3 | GPT-4o | 57.3 | #741 |
| 4 | Gemini 2.5 Pro | 57 | #262 |
| 5 | Yi 1.5 34B Chat | 53 | #2513 |
| 6 | Gemini 3 Pro (Preview) | 39 | #111 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=neo-classic-word-cloze-random-distractors · How It Works · Data refreshed daily, snapshot 2026-11-12.