List your company

Mathematical Reasoning on MathVerse (testmini)

71.5AccuracyPRISM + GSPO
57.87661.41364.9568.487Apr 30, 2026

Evaluation results

MethodLinks
PRISM + GSPOBase Model=Qwen3-VL-8B2026.0471.5
PRISM + GRPOBase Model=Qwen3-VL-8B2026.0471.3
PRISM + DAPOBase Model=Qwen3-VL-8B2026.0470.9
+ DAPOBase Model=Qwen3-VL-8B2026.0469.8
PRISM + GRPOBase Model=Qwen3-VL-4B2026.0468.6
PRISM + DAPOBase Model=Qwen3-VL-4B2026.0468.2
+ GRPOBase Model=Qwen3-VL-8B2026.0466.9
PRISM + GSPOBase Model=Qwen3-VL-4B2026.0466.6
+ GSPOBase Model=Qwen3-VL-8B2026.0465.5
+ DAPOBase Model=Qwen3-VL-4B2026.0465.1
+ GRPOBase Model=Qwen3-VL-4B2026.0464.5
+ GSPOBase Model=Qwen3-VL-4B2026.0464
InstructBase Model=Qwen3-VL-8B2026.0462.4
PRISMBase Model=Qwen3-VL-8B2026.0462.2
+ SFTBase Model=Qwen3-VL-8B2026.0460.4
PRISMBase Model=Qwen3-VL-4B2026.0459.5
InstructBase Model=Qwen3-VL-4B2026.0459
+ SFTBase Model=Qwen3-VL-4B2026.0458.4