Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Comprehensive Reasoning on ScienceQA
Loading...
84.2
Accuracy
Unsilencing Latent Reasoning
68.288
72.419
76.55
80.681
May 4, 2026
Accuracy
Updated 29d ago
Evaluation Results
Method
Method
Links
Accuracy
Unsilencing Latent Reasoning
Backbone=Qwen2.5VL-7B
2026.05
84.2
MCoT
Backbone=Qwen2.5VL-7B
2026.05
83.9
CCoT
Backbone=Qwen2.5VL-7B
2026.05
83.8
CoVT
Backbone=Qwen2.5VL-7B
2026.05
83.8
DMLR
Backbone=Qwen2.5VL-7B
2026.05
83.4
LVRRF
Backbone=Qwen2.5VL-7B
2026.05
83.1
LVR
Backbone=Qwen2.5VL-7B
2026.05
82.8
Vanilla
Backbone=Qwen2.5VL-7B
2026.05
82.3
Monet
Backbone=Qwen2.5VL-7B
2026.05
78.8
ICoT
Backbone=Qwen2.5VL-7B
2026.05
78.4
Unsilencing Latent Reasoning
Backbone=Qwen2.5VL-3B
2026.05
74.3
MCoT
Backbone=Qwen2.5VL-3B
2026.05
74.1
CCoT
Backbone=Qwen2.5VL-3B
2026.05
73.9
DMLR
Backbone=Qwen2.5VL-3B
2026.05
73.8
LVRRF
Backbone=Qwen2.5VL-3B
2026.05
73.7
Vanilla
Backbone=Qwen2.5VL-3B
2026.05
73.5
LVR
Backbone=Qwen2.5VL-3B
2026.05
73.2
ICoT
Backbone=Qwen2.5VL-3B
2026.05
68.9
Feedback
Search any
task
Search any
task