Arena-Hard
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Arena-Hard
42.7Win Rate
73
Arena-Hard
22.1Score
46
Arena-Hard v2
85.9Score
38
Arena-Hard v0.1
35.2Win Rate (%)
36
Arena-Hard v0.1
50Win Rate
31
Arena Hard v0.1
4.5Wrong Rate (%)
31
Arena-Hard Creative Writing v2
90.8Score
25
Arena-Hard Vanilla
57.5Instruction Following Score
19
Arena Hard
63.5Win Rate
18
Arena-Hard Style-Con
57.7Score
17
Arena-Hard
63.9Win Rate (%)
16
Arena-Hard
84Win Rate
16
Arena-Hard v2.0
33.9Win Rate
14
Arena-Hard v2
18.2Score
14
Arena-Hard
3.29Accepted Length (τ)
12
Arena-Hard 2.0 (test)
52.9Overall Accuracy
12
Arena-Hard (test)
12.61Error
12
Arena-Hard
5.17Error (%)
12
Arena-Hard
32.6Win Rate (%)
11
Arena Hard v2
79.9Score
10
Arena-Hard vs gpt-4-0314 (test)
57.6Win Rate
9
Arena-Hard V2
73.2Win Rate
9
Arena-Hard V2.0
7.03Win Rate
9
Arena-Hard v0.1
78.3Arena-Hard Score
9
Arena-Hard v2.0
47.8Score
8