List your company

Multimodal Geometry Reasoning on Geo3K

49.75AccuracyPair
31.102835.943940.78545.6261May 27, 2026

Evaluation results

MethodLinks
PairModel=Qwen3-VL-4B, Group Size G=162026.0549.751,830.3740.311,3882.57
PairModel=Qwen3-VL-4B, Group Size G=82026.0549.672,136.6729.96,9123.63
BPPOModel=Qwen3-VL-4B, Group Size G=82026.0549.432,063.2332.46,4903.86
CPPOModel=Qwen3-VL-4B, Group Size G=162026.0549.433,851.84-25.611,7732.49
BPPOModel=Qwen3-VL-4B, Group Size G=162026.0548.931,665.1345.710,7102.73
GRPOModel=Qwen3-VL-4B, Group Size G=162026.0548.873,066.45—29,2641
CPPOModel=Qwen3-VL-4B, Group Size G=82026.0548.224,073.86-33.67,7503.23
GRPOModel=Qwen3-VL-4B, Group Size G=82026.0548.123,049.95—25,0711
GRPO+FIRSTNModel=Qwen3-VL-4B, Group Size G=162026.0547.973,006.12227,0071.08
GRPO+FIRSTNModel=Qwen3-VL-4B, Group Size G=82026.0547.123,016.771.122,9811.09
Qwen3-VL-4BModel=Qwen3-VL-4B2026.0531.82————