Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Multi-Task Reasoning on MMLU-Pro
Loading...
69.3
Pass@1
Composition-RL
22.6872
34.7886
46.89
58.9914
Feb 12, 2026
Feb 26, 2026
Mar 13, 2026
Mar 28, 2026
Apr 11, 2026
Apr 26, 2026
May 11, 2026
Pass@1
Updated 22d ago
Evaluation Results
Method
Method
Links
Pass@1
Composition-RL
Backbone=Qwen3-14B-Base
2026.02
69.3
Standard RLVR
Backbone=Qwen3-14B-Base
2026.02
67.2
Composition-RL
Backbone=Qwen3-30B-A3B...
2026.02
64.6
Composition-RL
Backbone=Qwen3-4B-Base...
2026.02
64.5
Composition-RL
Backbone=Qwen3-8B-Base
2026.02
64.5
Composition-RL
Backbone=Qwen3-4B-Base...
2026.02
63.8
Standard RLVR
Backbone=Qwen3-8B-Base
2026.02
62.6
Standard RLVR
Backbone=Qwen3-30B-A3B...
2026.02
62.6
Composition-RL
Backbone=Qwen3-4B-Base
2026.02
61.4
Standard RLVR
Backbone=Qwen3-4B-Base
2026.02
58.6
Before RL (Base)
Backbone=Qwen2.5-7B, T...
2026.05
47.21
GT-Reward
Backbone=Qwen2.5-7B, T...
2026.05
43.17
VIGOR
Backbone=Qwen2.5-7B, T...
2026.05
43.09
INTUITOR
Backbone=Qwen2.5-7B, T...
2026.05
43.04
GT-Reward
Backbone=Qwen2.5-3B, T...
2026.05
38.17
Before RL (Base)
Backbone=Qwen2.5-3B, T...
2026.05
36.92
VIGOR
Backbone=Qwen2.5-3B, T...
2026.05
32.65
INTUITOR
Backbone=Qwen2.5-3B, T...
2026.05
24.48
Feedback
Search any
task
Search any
task