CORAL
Models retrieve passages, answer questions, and cite supporting evidence across conversations. Retrieval metrics, response-overlap metrics, model judgments, and citation precision and recall score different parts of the task.
The maker announced a second dataset version in May 2025. The paper's protocol and version must accompany results.
- Made by
- Renmin University of China, Beijing Academy of Artificial Intelligence, Huawei Poisson Lab, and Waseda University
- Released
- October 2024
- Kind
- Questions that need search
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Multi-turn
- Refresh
- New versions replace it
- Run by
- Anyone
- Last model added
- September 2024
- Scores refreshed
- November 20th, 2026
In the authors’ words
Retrieval-Augmented Generation (RAG) has become a powerful paradigm for enhancing large language models (LLMs) through external knowledge retrieval. Despite its widespread attention, existing academic research predominantly focuses on single-turn RAG, leaving a significant gap in addressing the complexities of multi-turn conversations found in real-world applications. To bridge this gap, we introduce CORAL, a large-scale benchmark designed to assess RAG systems in realistic multi-turn conversational settings. CORAL includes diverse information-seeking conversations automatically derived from Wikipedia and tackles key challenges such as open-domain coverage, knowledge intensity, free-form responses, and topic shifts. It supports three core tasks of conversational RAG: passage retrieval, response generation, and citation labeling. We propose a unified framework to standardize various conversational RAG methods and conduct a comprehensive evaluation of these methods on CORAL, demonstrating substantial opportunities for improving existing approaches.
KD-ANCE-C tops the leaderboard with 23.2 Qwen2.5-1.5B leads with 16.3
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper gives no combined score, so this is mean reciprocal rank on its first task, conversational passage retrieval. 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. The best model we track is Qwen2.5-1.5B, at 16.3. This view shows only models we track, each at its best.
- 1 KD-ANCE-C October 2024 CORAL test, CDR 23.2
- 2 Qwen2.5-1.5B-SFT October 2024 CORAL test, CQR 23.1
- 3 KD-ANCE-Q October 2024 CORAL test, CDR 22.6
- 4 GPT-3.5 (snapshot unspecified) October 2024 CORAL test, CQR 21.2
- 5 Conv-ANCE-C October 2024 CORAL test, CDR 20.5
- 6 Conv-ANCE-Q October 2024 CORAL test, CDR 19.8
- 7
Qwen2.5-1.5B October 2024 CORAL test, CQR 16.3
Source: The CORAL paper
Source: The CORAL paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | KD-ANCE-C | 23.2 | October 2024 CORAL test, CDR | Official leaderboard | |
| 2 | Qwen2.5-1.5B-SFT | 23.1 | October 2024 CORAL test, CQR | Official leaderboard | |
| 3 | KD-ANCE-Q | 22.6 | October 2024 CORAL test, CDR | Official leaderboard | |
| 4 | GPT-3.5 (snapshot unspecified) | 21.2 | October 2024 CORAL test, CQR, LLM4CS | Official leaderboard | |
| 5 | Conv-ANCE-C | 20.5 | October 2024 CORAL test, CDR | Official leaderboard | |
| 6 | Conv-ANCE-Q | 19.8 | October 2024 CORAL test, CDR | Official leaderboard | |
| 7 | | 16.3 | October 2024 CORAL test, CQR | Sep 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 16.3 | October 2024 CORAL test, CQR | Sep 2024 | Official leaderboard |
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official CORAL 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 Knowledge
- MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects.
Intern-S2-397B 89.8%
- MMLU In use 57-subject multiple-choice exam questions from elementary to professional level. Scored by accuracy.
Seed1.8 92.3%
- SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy.
Qwen 3.6 Max (Preview) 73.9%
- WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1.
Tencent Hy4 preview 83.9%
- BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly.
Gemma 4 31B IT 74.4%
- HLE-Diamond In use Models answer 1,000 cleaned expert questions from Humanity's Last Exam, 500 on reasoning and 500 on knowledge, closed book. Graded on exact answers. Claude Fable 5.2 77.9%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in September 2024, for Qwen2.5-1.5B. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (7 entries).
- Pass Models scored 1 model we track, plus 6 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-1.5B.
- Note Frontier models None of its models are on the capabilities index.
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 1 row for models we track and 6 other entries at the last read.
- Pass Last read November 20th, 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.