MT-Bench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
MT-Bench GPT-4 evaluator (test)
6.46MT-Bench Score
12
MT-bench (test)
2.78Speedup Ratio (γ)
12
MT-Bench GPT-4o judge (test)
7.76MT-Bench Score
12
MT Bench
0.494Kendall's Tau
11
MT-bench
24.1Throughput (TPS)
10
MT-Bench short-context
8.13MT-Bench Score
10
MT-Bench (test)
7.1MT-Bench Mean Score
9
MT-Bench
83.69Human Agreement Rate
9
MT-Bench
6.55MT-Bench Delta Score
9
MT-bench
47.1Throughput (TPS)
9
MT-Bench
8.32Latency (ms/tok)
8
MT-Bench
8.94Latency (ms/tok)
8
MT-Bench
3.61Acceptance Length
8
MT-Bench single-turn
83.9MT-Bench Score
8
MT-Bench (hold-out)
9.6Score
8
MT-Bench (MTB)
2.53Speedup Factor
8
MT-Bench 101
3.9Likert Score
8
MT-Bench (test)
5.71Turn 1 Score
8
MT-Bench OOD (test)
73Score
8
MT-bench First Turn
97Agreement
8
MT-Bench zh
6.66Overall Score
7
MT-Bench
8.54Accuracy
7
MT-Bench
8.03Overall Score
7
MT Bench 90 prompts (test)
808Total Wins
7
MT-Bench English
8.73Score
6