Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Math on AIME24 (Avg@16)
Loading...
76.5
Avg@16
Base
28.0152
40.6026
53.19
65.7774
May 8, 2026
May 13, 2026
May 18, 2026
May 23, 2026
May 28, 2026
Jun 2, 2026
Jun 7, 2026
Avg@16
Updated 1mo ago
Evaluation Results
Method
Method
Links
Avg@16
Base
Backbone=Qwen3-8B (w/...
2026.06
76.5
+Forward KL w/ Clip
Backbone=Qwen3-8B (w/...
2026.06
76.5
TRD
Backbone=Qwen3-8B (w/...
2026.06
76.5
+Reverse KL w/ Top-K
Backbone=Qwen3-8B (w/...
2026.06
75.6
+Reverse KL
Backbone=Qwen3-8B (w/...
2026.06
75.4
+Forward KL
Backbone=Qwen3-8B (w/...
2026.06
74.8
Base
Backbone=Qwen3-4B-Inst...
2026.06
63.1
TRD
Backbone=Qwen3-4B-Inst...
2026.06
63.1
+Reverse KL w/ Top-K
Backbone=Qwen3-4B-Inst...
2026.06
63
+Forward KL w/ Clip
Backbone=Qwen3-4B-Inst...
2026.06
62.1
+Forward KL
Backbone=Qwen3-4B-Inst...
2026.06
60.3
+Reverse KL
Backbone=Qwen3-4B-Inst...
2026.06
58.4
AIPO
Model=Qwen2.5-32B-Inst...
2026.05
39.06
GRPO
Model=Qwen2.5-32B-Inst...
2026.05
35.71
SFT
Model=Qwen2.5-32B-Inst...
2026.05
32.61
Original
Model=Qwen2.5-32B-Inst...
2026.05
29.88
Feedback
Search any
task
Search any
task