The DOI matches its DataCite record; 59 of the 59 references it cites were found in a registry or an index; 4 of the 29 sources its claims cite were checked for support.
Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena
Open accessAbstract
Evaluating large language model (LLM) based chat assistants is challenging due to their broad capabilities and the inadequacy of existing benchmarks in measuring human preferences. To address this, we explore using strong LLMs as judges to evaluate these models on more open-ended questions. We examine the usage and limitations of LLM-as-a-judge, including position, verbosity, and self-enhancement biases, as well as limited reasoning ability, and propose solutions to mitigate some of them. We then verify the agreement between LLM judges and human preferences by introducing two benchmarks: MT-bench, a multi-turn question set; and Chatbot Arena, a crowdsourced battle platform. Our results reveal that strong LLM judges like GPT-4 can match both controlled and crowdsourced human preferences well, achieving over 80% agreement, the same level of agreement between humans. Hence, LLM-as-a-judge is a scalable and explainable way to approximate human preferences, which are otherwise very expensive to obtain. Additionally, we show our benchmark and traditional benchmarks complement each other by evaluating several variants of LLaMA and Vicuna. The MT-bench questions, 3K expert votes, and 30K conversations with human preferences are publicly available at https://github.com/lm-sys/FastChat/tree/main/fastchat/llm_judge.
Overview
CoChat Summary
TL;DR
This paper examines whether strong LLMs can serve as judges for chatbot responses by comparing their judgments with human preferences on two new benchmarks, MT-bench and Chatbot Arena. It finds that GPT-4 can reach over 80% agreement with human preferences, comparable to human-human agreement, while also documenting several judge biases and mitigation strategies.
Key findings
- GPT-4’s judgments aligned closely with human experts on MT-bench: 85% agreement without ties, versus 81% agreement among humans; on Chatbot Arena, GPT-4 also showed high agreement with crowd votes, above 80%.
- Position bias was substantial: with the default prompt, consistency was 65.0% for GPT-4, 46.2% for GPT-3.5, and 23.8% for Claude-v1; swapping positions and few-shot examples raised GPT-4 consistency to 77.5%.
- Verbosity bias appeared in a “repetitive list” attack: Claude-v1 and GPT-3.5 failed 91.3% of cases, while GPT-4 failed 8.7%.
- Reference-guided grading reduced math-judging failures from 14/20 under the default prompt to 3/20, with chain-of-thought at 6/20; self-enhancement was suggested by GPT-4 favoring itself with a 10% higher win rate and Claude-v1 with a 25% higher win rate, but the study could not confirm the bias.
Methods
- The authors created MT-bench, 80 multi-turn questions across 8 categories with 10 questions each, and Chatbot Arena, a crowdsourced anonymous battle platform; they collected about 3K expert votes from 58 labelers on MT-bench and about 30K Arena votes, sampling 3K for analysis.
- They compared LLM judges—mainly GPT-4, GPT-3.5, and Claude-v1—using pairwise comparison, single-answer grading, and reference-guided grading against human expert and crowd votes, measuring agreement and win rates.
- They also evaluated LLaMA and Vicuna variants on MMLU, TruthfulQA, and MT-bench with GPT-4 single-answer grading to test how preference-based and capability-based benchmarks complement each other.
Limitations
- The paper states that it emphasizes helpfulness and largely neglects safety, honesty, and harmlessness; it also notes that helpfulness dimensions such as accuracy, relevance, and creativity are combined into one metric, and its bias mitigations are preliminary.
- The paper reports that self-enhancement bias could not be determined because of limited data and small differences, and that its position-bias test is difficult because the compared answers are very similar and sometimes indistinguishable even to humans.
- The text makes evident that few-shot judging may introduce new biases and makes API calls about 4× more expensive, while zero-shot Vicuna-13B as a judge had low consistency (11.2–16.2%) and high error rates (22.5–78.8%), limiting its use as a scalable open-source judge.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
3 Valid · 2 Not valid · 8 Not checked · 98 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Once aligned with humans, these chat models are strongly preferred by human users over the original, unaligned models on which they are built.”
§ Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena > 1 Introduction
What the sources say
- Not checked — no full text of the sourcecites [31] source
- Partly supportedcites [2] sourceExperimental — accuracy not yet measured
Section 4.5 (p. 21) says crowdworkers clearly preferred the online RLHF models in comparisons with context-distilled and earlier RLHF models. But those preference comparisons do not establish that users strongly preferred them over the original, unaligned base models; the paper discusses base-model improvements on other evaluations (Introduction, p. 4). (LLM, conf=0.86)
- Not matchedcites [30]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [8]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [52]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [48]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [14]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We show that some of the biases are minor or can be mitigated.”
§ Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena > GPT-4 Judgment:
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Note that in math/coding category, GPT-3.5 and GPT-4 have similar overall win-rate because they both failed to answer some hard questions, but GPT-4 is still significantly better than GPT-3 in the direct pairwise comparison or single-answer grading.”
§ Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena > 4.3 Win rates under different judges
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“On the other hand, a small high-quality conversation dataset can quickly teach the model a style preferred by GPT-4 (or approximately human) but cannot improve MMLU significantly, as shown by the Vicuna-7B (selected) which is trained with only 4.8M tokens or 3K conversations.”
§ Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena > 5 Human Preference Benchmark and Standardized Benchmark
plainstates a causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Most LLM judges favor the first position.”
§ Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena > 3.3 Limitations of LLM-as-a-Judge
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
60 more claims
Create a free account to see all 111 claims and every source verdict.
- Probably not a substantial claim. 22
- It may state two separate claims as one. 29
- The quote only partly states this claim. 2
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source doesn’t support the claim as written.
In the citing paper
“To mitigate stochastic variance and”
From Beyond the Context Window: A Cost-Performance Analysis of Fact-Based Memory vs. Long-Context LLMs for Persistent Agents (2026) · cites this paper as Zheng et al., 2023
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe cited source doesn’t support the claim as written.
In the citing paper
“Others employ the LLM-as-a-judge methodology [36], yet these efforts lack both a comprehensive framework design and verification of human consistency.”
From DeepResearch Bench: A Comprehensive Benchmark for Deep Research Agents (2025) · cites this paper as [36]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe cited source doesn’t support the claim as written.
In the citing paper
“LLM-based evaluators are mainly used in well-defined tasks that require simple information extraction and classification, which has been shown to have high precision Zheng et al. (2023).”
From TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks (2024) · cites this paper as Zheng et al. (2023)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The choice of judge can shift accuracy by 5-10 percentage points on identical retrieval output (Zheng et al., 2023).”
From SmartSearch: How Ranking Beats Structure for Conversational Memory Retrieval (2026) · cites this paper as Zheng et al., 2023
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“considering that Alpaca is known to perform worse than our other open-source autoregressive models, Vicuna and Llama-2 (refs. 21,35)”
From Adapted large language models can outperform medical experts in clinical text summarization (2024) · cites this paper as [21]
Experimental — accuracy not yet measured
106 more claims about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
BLEU: a method for automatic evaluation of machine translation
Papineni, K., Roukos, S., Ward, T., & Zhu, W.-J. (2001) · Proceedings of the 40th Annual Meeting on Association for Computational Linguistics - ACL '02 · doi:10.3115/1073083.1073135 · Cited by 21,344 · cited 2×
“Traditional evaluation metrics based on the similarity between outputs and reference answers (e.g., ROUGE [25], BLEU [32]) are also ineffective for these questions.”
MethodLLaMA: Open and Efficient Foundation Language Models
Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al. (2023) · arXiv · doi:10.48550/arxiv.2302.13971 · Cited by 4,602 · cited 6×
“5, Claude-V1, Vicuna-13B, Alpaca-13B [38], and LLaMA-13B [39].”
BackgroundInfluentialChain-Of-Thought Prompting Elicits Reasoning in Large Language Models
Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q. V., & Zhou, D. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-1800 · Cited by 4,505 · cited 1×
“Chain-of-thought is a widely used technique to improve LLM’s reasoning capability [39].”
MethodChain-of-Thought Prompting Elicits Reasoning in Large Language Models
Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q., & Zhou, D. (2022) · arXiv · doi:10.48550/arxiv.2201.11903 · Cited by 4,263 · cited 1×
“Chain-of-thought is a widely used technique to improve LLM’s reasoning capability [39].”
MethodQLoRA: Efficient Finetuning of Quantized LLMs
Dettmers, T., Pagnoni, A., Holtzman, A., & Zettlemoyer, L. (2023) · arXiv · doi:10.48550/arxiv.2305.14314 · Cited by 3,832 · cited 1×
“This approach has been tried in our earlier blog post [7] and other concurrent or follow-up work [5, 25, 12, 11, 43, 15].”
BackgroundTraining Language Models to Follow Instructions with Human Feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-2011 · Cited by 3,117 · cited 2×
“There has been a proliferation of LLM-based chat assistants (chatbots) that leverage supervised instruction fine-tuning and reinforcement learning with human feedback (RLHF) to unlock new instruction following and conversational abilities [27, 2, 26, 7, 43].”
BackgroundHellaSwag: Can a Machine Really Finish Your Sentence?Preprint
Zellers, R., Holtzman, A., Bisk, Y., Farhadi, A., & Choi, Y. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.07830 · Cited by 2,074 · cited 2×
“• Core-knowledge benchmarks, including MMLU [16], HellaSwag [41], ARC [8], WinoGrande [31], HumanEval [6], GSM-8K [9], and AGIEval [42], evaluate the core capabilities of pre-trained LLMs using zero-shot and few-shot benchmark sets.”
MethodSparks of Artificial General Intelligence: Early experiments with GPT-4
Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee, P., Lee, Y. T., Li, Y., Lundberg, S., et al. (2023) · arXiv · doi:10.48550/arxiv.2303.12712 · Cited by 1,526 · cited 2×
“This approach has been tried in our earlier blog post [7] and other concurrent or follow-up work [5, 25, 12, 11, 43, 15].”
BackgroundEvaluating Large Language Models Trained on Code
Chen, M., Tworek, J., Jun, H., Yuan, Q., Pinto, H. P. D. O., Kaplan, J., Edwards, H., Burda, Y., Joseph, N., Brockman, G., et al. (2021) · arXiv · doi:10.48550/arxiv.2107.03374 · Cited by 1,436 · cited 2×
“• Core-knowledge benchmarks, including MMLU [16], HellaSwag [41], ARC [8], WinoGrande [31], HumanEval [6], GSM-8K [9], and AGIEval [42], evaluate the core capabilities of pre-trained LLMs using zero-shot and few-shot benchmark sets.”
MethodWizardLM: Empowering Large Pre-Trained Language Models to Follow Complex InstructionsPreprint
Xu, C., Sun, Q., Zheng, K., Geng, X.-B., Zhao, P., Feng, J., Tao, C., Lin, Q.-W., & Jiang, D. (2023) · International Conference on Learning Representations · doi:10.48550/arxiv.2304.12244 · Cited by 1,373 · cited 1×
“There has been a proliferation of LLM-based chat assistants (chatbots) that leverage supervised instruction fine-tuning and reinforcement learning with human feedback (RLHF) to unlock new instruction following and conversational abilities [30, 2, 29, 8, 51, 47, 14].”
Background
Research works printed but not linked: 2
- Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pages 74-81, 2004.
- Zongheng Yang, Zhanghao Wu, Michael Luo, Wei-Lin Chiang, Romil Bhardwaj, Woosuk Kwon, Siyuan Zhuang, Frank Sifei Luan, Gautam Mittal, Scott Shenker, and Ion Stoica. SkyPilot: An intercloud broker for sky computing. In 20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23), pages 437-455, Boston, MA, April 2023. USENIX Association.
Cited by
A Survey on Evaluation of Large Language Models
Chang, Y., Wang, X., Wang, J., Wu, Y., Yang, L., Zhu, K., Chen, H., Yi, X., Wang, C., Wang, Y., et al. (2024) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3641289 · Cited by 4,254
“Moreover, MT-Bench effectively overcomes the limitations in traditional evaluation approaches, particularly in gauging a model’s competence in handling intricate multi-turn dialogue inquiries.”
BackgroundInfluentialG-Eval: NLG Evaluation using Gpt-4 with Better Human Alignment
Liu, Y., Iter, D., Xu, Y., Wang, S., Xu, R., & Zhu, C. (2023) · Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2023.emnlp-main.153 · Cited by 948
“Meanwhile, it is becoming popular to use more powerful LLMs like GPT-4 to evaluate smaller or student models, like in Alpaca (Taori et al., 2023) and Vicuna (Zheng et al., 2023).”
MethodG-Eval: NLG Evaluation using GPT-4 with Better Human Alignment
Liu, Y., Iter, D., Xu, Y., Wang, S., Xu, R., & Zhu, C. (2023) · arXiv · doi:10.48550/arxiv.2303.16634 · Cited by 948
“Meanwhile, it is becoming popular to use more powerful LLMs like GPT-4 to evaluate smaller or student models, like in Alpaca (Taori et al., 2023) and Vicuna (Zheng et al., 2023).”
MethodAdapted large language models can outperform medical experts in clinical text summarization
Van Veen, D., Van Uden, C., Blankemeier, L., Delbrouck, J.-B., Aali, A., Bluethgen, C., Pareek, A., Polacin, M., Reis, E. P., Seehofnerová, A., et al. (2024) · Nature Medicine · doi:10.1038/s41591-024-02855-5 · Cited by 891
Intern VL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks
Chen, Z., Wu, J., Wang, W., Su, W., Chen, G., Xing, S., Zhong, M., Zhang, Q., Zhu, X., Lu, L., et al. (2024) · 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52733.2024.02283 · Cited by 734
“Due to the aligned feature space with LLMs, our model can smoothly integrate with existing LLMs, such as LLaMA series [106, 107], Vicuna [145], and InternLM [104].”
MethodInfluential
Metadata
- Authors
- Zheng, L., Chiang, W.-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E. P., Zhang, H., Gonzalez, J. E., & Stoica, I.
- Venue
- arXiv
- Publisher
- arXiv
- Published
- 2023
- Pages
- 46595-46623
- ISSN
- 1049-5258 (ISSN-L)
- arXiv
- 2306.05685
- OpenAlex
- W7133239970
- Cited by
- 1,888
- References
- 2
- Citations by year
- 2026: 1115 · 2025: 413 · 2024: 308 · 2023: 51
- FWCI
- 2548.93
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- neural information processing systems (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- VMware
- Samsung
- Topics
- Social Sciences › Law › Artificial Intelligence in Law
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Keywords
- large language models, LLM-as-a-judge, automated assessment, MT-Bench, Chatbot Arena, benchmarking, chatbots
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena. https://app.cochat.ai/papers/10.48550/arxiv.2306.05685
- The paper
- Zheng, L., Chiang, W.-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E. P., Zhang, H., Gonzalez, J. E., & Stoica, I. (2023). Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena. arXiv. https://doi.org/10.48550/arxiv.2306.05685
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.