Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Human Preference Alignment on Arena-Hard
Loading...
63.9
Win Rate (%)
KTO
8.26
22.705
37.15
51.595
Jun 8, 2026
Win Rate (%)
Updated 1mo ago
Evaluation Results
Method
Method
Links
Win Rate (%)
KTO
Backbone=Qwen3-4B-Base
2026.06
63.9
RePO
Backbone=Qwen3-4B-Base
2026.06
60.1
RePO_det
Backbone=Qwen3-4B-Base
2026.06
59.9
IPO
Backbone=Qwen3-4B-Base
2026.06
47.8
DPO
Backbone=Qwen3-4B-Base
2026.06
44.5
RPO
Backbone=Qwen3-4B-Base
2026.06
41.9
TDPO
Backbone=Qwen3-4B-Base
2026.06
30.9
KTO
Backbone=Qwen3-1.7B-Base
2026.06
30.4
RePO
Backbone=Qwen3-1.7B-Base
2026.06
27.1
RePO_det
Backbone=Qwen3-1.7B-Base
2026.06
26.6
Base
Backbone=Qwen3-4B-Base
2026.06
25.4
DPO
Backbone=Qwen3-1.7B-Base
2026.06
23.4
IPO
Backbone=Qwen3-1.7B-Base
2026.06
21.9
RPO
Backbone=Qwen3-1.7B-Base
2026.06
17.7
TDPO
Backbone=Qwen3-1.7B-Base
2026.06
14.5
Base
Backbone=Qwen3-1.7B-Base
2026.06
10.4
Feedback
Search any
task
Search any
task