Phi-4-Reasoning-Vision-15B Benchmark Details

Phi-4-Reasoning-Vision-15B currently shows benchmark results led by MathVista (testmini) (1 / 2, score 74.10), MMStar (1 / 2, score 63.90), HallusionBench (1 / 2, score 63.90).

Benchmark Results

Phi-4-Reasoning-Vision-15B

Benchmark Results

Visual Understanding

12 evaluations

Benchmark / mode

Score

Rank/total

MathVista (testmini)

Standard Mode

68.70

2 / 2

MathVista (testmini)

Thinking Mode

74.10

1 / 2

MMStar

Standard Mode

63.30

2 / 2

MMStar

Thinking Mode

63.90

1 / 2

HallusionBench

Standard Mode

63.10

2 / 2

HallusionBench

Thinking Mode

63.90

1 / 2

MMMU

Standard Mode

66 / 76

MMMU

Thinking Mode

64 / 76

MathVerse (testmini)

Standard Mode

43.80

2 / 2

MathVerse (testmini)

Thinking Mode

53.10

1 / 2

MathVision (testmini)

Standard Mode

34.20

2 / 2

MathVision (testmini)

Thinking Mode

36.20

1 / 2

GUI Grounding

2 evaluations

Benchmark / mode

Score

Rank/total

ScreenSpot-v2

Standard Mode

88.30

1 / 2

ScreenSpot-v2

Thinking Mode

88.10

2 / 2

Documents & Charts

6 evaluations

Benchmark / mode

Score

Rank/total

AI2D

Standard Mode

84.70

1 / 2

AI2D

Thinking Mode

79.70

2 / 2

ChartQA

Standard Mode

76.50

2 / 2

ChartQA

Thinking Mode

82.90

1 / 2

OCRBench

Standard Mode

75.60

1 / 2

OCRBench

Thinking Mode

73.70

2 / 2

Compare with other models