chat benchmark
MT-Bench Leaderboard
MT-Bench is a challenging multi-turn benchmark for evaluating large language models' ability to engage in coherent, informative, and engaging conversations.
Updated Sep 5, 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 score93.50% | Percentile100.00% | Participants12 | EvidenceC | Evaluated |
| Rank02 | ModelNV | Normalized score91.70% | Percentile90.91% | Participants12 | EvidenceC | Evaluated |
| Rank03 | ModelDE | Normalized score90.20% | Percentile81.82% | Participants12 | EvidenceC | Evaluated |
| Rank04 | ModelNR | Normalized score89.90% | Percentile72.73% | Participants12 | EvidenceC | Evaluated |
| Rank05 | ModelAC | Normalized score87.50% | Percentile63.64% | Participants12 | EvidenceC | Evaluated |
| Rank06 | ModelMA | Normalized score86.30% | Percentile54.55% | Participants12 | EvidenceC | Evaluated |
| Rank07 | ModelAC | Normalized score84.10% | Percentile45.45% | Participants12 | EvidenceC | Evaluated |
| Rank08 | ModelMA | Normalized score83.50% | Percentile36.36% | Participants12 | EvidenceC | Evaluated |
| Rank09 | ModelMA | Normalized score83.00% | Percentile27.27% | Participants12 | EvidenceC | Evaluated |
| Rank10 | ModelNV | Normalized score81.00% | Percentile18.18% | Participants12 | EvidenceC | Evaluated |
| Rank11 | ModelMA | Normalized score76.80% | Percentile9.09% | Participants12 | EvidenceC | Evaluated |
| Rank12 | ModelNV | Normalized score8.99% | Percentile0.00% | 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.
What is MT-Bench?
What MT-Bench measures and how its scores work.
MT-Bench is a multi-turn dialogue benchmark that uses strong LLMs as judges for scalable and explainable evaluation.
It measures multi-turn dialogue capabilities using a Normalized score expressed as a ratio.
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
- chat
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- overall
LLM Stats. 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 93.50%.
What are the top three models on MT-Bench?
The current leaders are Qwen2.5 72B Instruct (93.50%), Llama-3.3 Nemotron Super 49B v1 (91.70%), and DeepSeek-V2.5 (90.20%).
Which MT-Bench model has the lowest official input price?
Pixtral-12B has the lowest matched official input price at $0.15 input / $0.15 output per 1M tokens.
Which models are fastest among MT-Bench results?
The fastest matched records are Mistral Small 3 24B Instruct (134.00 tok/s via Mistral AI), DeepSeek-V2.5 (100.00 tok/s via DeepSeek), and Mistral Large 2 (42.00 tok/s via Google).
Does the highest normalized score result prove overall model quality?
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.