reasoning benchmark
MT-Bench Leaderboard
MT-Bench is a challenging multi-turn benchmark that measures the ability of large language models to engage in coherent, informative, and engaging conversations. It uses strong LLMs as judges for scalable and explainable evaluation of multi-turn dialogue capabilities.
Updated Aug 17, 2026
Models12
Model coverage12
MetricNormalized score
EvidenceB
MT-Bench Ranking
Higher normalized score ranks better on this benchmark.
Sort by
Rank | Model | Normalized score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelAC | Normalized score93.5% | Percentile100.0% | Participants12 | EvidenceC | Evaluated |
| Rank02 | ModelNV | Normalized score91.7% | Percentile90.9% | Participants12 | EvidenceC | Evaluated |
| Rank03 | ModelDE | Normalized score90.2% | Percentile81.8% | Participants12 | EvidenceC | Evaluated |
| Rank04 | ModelNR | Normalized score89.9% | Percentile72.7% | Participants12 | EvidenceC | Evaluated |
| Rank05 | ModelAC | Normalized score87.5% | Percentile63.6% | Participants12 | EvidenceC | Evaluated |
| Rank06 | ModelMA | Normalized score86.3% | Percentile54.5% | Participants12 | EvidenceC | Evaluated |
| Rank07 | ModelAC | Normalized score84.1% | Percentile45.5% | Participants12 | EvidenceC | Evaluated |
| Rank08 | ModelMA | Normalized score83.5% | Percentile36.4% | Participants12 | EvidenceC | Evaluated |
| Rank09 | ModelMA | Normalized score83.0% | Percentile27.3% | Participants12 | EvidenceC | Evaluated |
| Rank10 | ModelNV | Normalized score81.0% | Percentile18.2% | Participants12 | EvidenceC | Evaluated |
| Rank11 | ModelMA | Normalized score76.8% | Percentile9.1% | Participants12 | EvidenceC | Evaluated |
| Rank12 | ModelNV | Normalized score9.0% | Percentile0.0% | Participants12 | EvidenceC | Evaluated |
MT-Bench Highlights
The leading models and scores on this benchmark.
MT-Bench Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for MT-Bench
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
Ranking basisThis mt-bench AI model leaderboard uses descending normalized score in the benchmark's original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
- 01ACQwen2.5 72B InstructAlibaba Cloud / Qwen Team
- Normalized score
- 93.5%
- Price
- $1.4 input / $5.6 output per 1M tokens
- Speed
- Up to 100 tok/s via Hyperbolic
Strengths
- Ranks #1 of 12 compared models
- 100th percentile on this benchmark
- C evidence result
Considerations
- This result measures MT-Bench, not total model capability
- 02NV
- Normalized score
- 91.7%
Strengths
- Ranks #2 of 12 compared models
- 91th percentile on this benchmark
- C evidence result
Considerations
- This result measures MT-Bench, not total model capability
- 03DEDeepSeek-V2.5DeepSeek
- Normalized score
- 90.2%
- Speed
- Up to 100 tok/s via DeepSeek
Strengths
- Ranks #3 of 12 compared models
- 82th percentile on this benchmark
- C evidence result
Considerations
- This result measures MT-Bench, not total model capability
- 04NRHermes 3 70BNous Research
- Normalized score
- 89.9%
Strengths
- Ranks #4 of 12 compared models
- 73th percentile on this benchmark
- C evidence result
Considerations
- This result measures MT-Bench, not total model capability
- 05ACQwen2.5 7B InstructAlibaba Cloud / Qwen Team
- Normalized score
- 87.5%
- Price
- $0.17 input / $0.70 output per 1M tokens
- Speed
- Up to 138 tok/s via Together
Strengths
- Ranks #5 of 12 compared models
- 64th percentile on this benchmark
- C evidence result
Considerations
- This result measures MT-Bench, not total model capability
Selection summary
Best AI Models for MT-Bench
Qwen2.5 72B Instruct currently leads MT-Bench with 93.5%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price and runtime.
Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.
What is MT-Bench?
What MT-Bench measures and how its scores work.
MT-Bench is a challenging multi-turn benchmark that measures the ability of large language models to engage in coherent, informative, and engaging conversations. It uses strong LLMs as judges for scalable and explainable evaluation of multi-turn dialogue capabilities.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- MT-Bench
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- mt-bench|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about MT-Bench.
Which model scores highest on MT-Bench?
Qwen2.5 72B Instruct is currently ranked first with 93.5%.
What does MT-Bench measure?
MT-Bench is a challenging multi-turn benchmark that measures the ability of large language models to engage in coherent, informative, and engaging conversations. It uses strong LLMs as judges for scalable and explainable evaluation of multi-turn dialogue capabilities.
Is a higher normalized score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
12 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.