OCNLI
Models decide whether a Chinese hypothesis follows from, contradicts, or remains neutral toward a premise. Accuracy measures agreement with the majority annotation.
The dataset uses original Chinese text rather than translations. Test labels are withheld, and examples without annotator agreement are excluded.
- Made by
- Indiana University, Allen Institute for AI and Chinese Language Understanding Evaluation (CLUE)
- Released
- October 2020
- Kind
- Reasoning in many languages
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- Public, plus a private test set
- Measures
- Capability
- Language
- Chinese
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- A host
In the authors’ words
Despite the tremendous recent progress on natural language inference (NLI), driven largely by large-scale investment in new datasets (e.g., SNLI, MNLI) and advances in modeling, most progress has been limited to English due to a lack of reliable datasets for most of the world's languages. In this paper, we present the first large-scale NLI dataset (consisting of ~56,000 annotated sentence pairs) for Chinese called the Original Chinese Natural Language Inference dataset (OCNLI). Unlike recent attempts at extending NLI to other languages, our dataset does not rely on any automatic translation or non-expert annotation. Instead, we elicit annotations from native speakers specializing in linguistics. We follow closely the annotation protocol used for MNLI, but create new strategies for eliciting diverse hypotheses. We establish several baseline results on our dataset using state-of-the-art pre-trained models for Chinese, and find even the best performing models to be far outpaced by human performance (~12% absolute performance gap), making it a challenging new resource that we hope will help to accelerate progress in Chinese NLU. To the best of our knowledge, this is the first human-elicited MNLI-style corpus for a non-English language.
RoBERTa (OCNLI trained) tops the leaderboard with 78.2%
Best score per entry from the benchmark's paper, read on November 19th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.
- 1 RoBERTa (OCNLI trained) 78.2%
- 2 BERT (OCNLI trained) 72.2%
- 3 ESIM (OCNLI trained) 59.8%
- 4 biLSTM (OCNLI trained) 59.2%
- 5 CBOW (OCNLI trained) 55.7%
Source: The OCNLI paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | RoBERTa (OCNLI trained) | 78.2% | Official leaderboard | ||
| 2 | BERT (OCNLI trained) | 72.2% | Official leaderboard | ||
| 3 | ESIM (OCNLI trained) | 59.8% | Official leaderboard | ||
| 4 | biLSTM (OCNLI trained) | 59.2% | Official leaderboard | ||
| 5 | CBOW (OCNLI trained) | 55.7% | Official leaderboard |
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Multilingual
- MMMLU In use MMLU professionally translated into 14 languages. Multiple-choice accuracy. Claude Mythos Preview 92.7%
- WMT24++ In use English-to-55-language translation of WMT24 test data. Scored with automatic metrics such as MetricX or xCOMET.
K-EXAONE 90.5%
- MMLU-ProX In use MMLU-Pro translated into up to 29 languages with expert review. Multiple-choice accuracy per language.
Qwen3.7-Max 87%
- CLIcK In use Models answer 1,995 multiple-choice questions on Korean language and culture. Questions come from official Korean exams and textbooks in eleven categories.
Solar Open 2 90.7%
- Global MMLU Lite 3.0 In use Models answer 400 human-translated MMLU questions per language in 23 languages. Version 3.0 adds Czech, Hungarian, Oriya, Slovak and Tajik and updates Italian. Accuracy. Claude Opus 5 95.7%
- MMMB In use Models answer multiple-choice questions about images in six languages across 15 categories. The set has about 12,000 questions. LFM2.5-VL-3B 83%
Data audit
All 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access Public tasks, plus a private test set.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the benchmark’s paper (5 entries).
- Pass Models scored 5 entries on its leaderboard, none a model we track.
- Note Newest score No scores yet.
- Note Frontier models No models scored yet.
- Note Headroom No scores yet.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 5 entries at the last read, none a model we track.
- Pass Last read November 19th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.