Benchmark
MT-Bench
MT-Bench is a challenging multi-turn benchmark that measures the ability of large language models to engage in coherent, informative, and engaging…
- Modality
- text
- Categories
- reasoning, roleplay, general, communication, creativity
- Openness
- unknown
- Source
- llm_stats
- Reported scores
- 12
Reported scores
llm_stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| DeepSeek-V2.5 | DeepSeek | 0.902 | 2024-05-08 | evidence |
| Hermes 3 70B | Nous Research | 8.99 | 2024-08-15 | evidence |
| Llama 3.1 Nemotron 70B Instruct | NVIDIA | 0.0899 | 2024-10-01 | evidence |
| Llama 3.1 Nemotron Nano 8B V1 | NVIDIA | 0.81 | 2025-03-18 | evidence |
| Llama-3.3 Nemotron Super 49B v1 | NVIDIA | 0.917 | 2025-03-18 | evidence |
| Ministral 8B Instruct | Mistral | 0.83 | 2024-10-16 | evidence |
| Mistral Large 2 | Mistral | 0.863 | 2024-07-24 | evidence |
| Mistral Small 3 24B Instruct | Mistral | 0.835 | 2025-01-30 | evidence |
| Pixtral-12B | Mistral | 0.768 | 2024-09-17 | evidence |
| Qwen2 7B Instruct | Qwen | 0.841 | 2024-07-23 | evidence |
| Qwen2.5 72B Instruct | Qwen | 0.935 | 2024-09-19 | evidence |
| Qwen2.5 7B Instruct | Qwen | 0.875 | 2024-09-19 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.