SOTA Medical Visual Question Answering on OmniMedVQA (test) and PapersWithCode

Method
M3LLM-8B 2025.11	85.1	85.5	89.3	90.2	79.8	83.6	88.7	78.2	85.7	-	-	-	-	-	-
MedLVR (ours) 2026.04	80.4	-	-	80.2	-	-	77.5	51.8	78	82	-	-	76.6	81.9	70.2
InternVL3-8B 2025.11	78.9	87.5	80.4	77.2	81.8	82.2	87.3	76.4	79	-	-	-	-	-	-
Lingshu-7B 2025.11	77.2	88.9	83.3	89.9	83.2	82.7	86.4	80.6	82.8	-	-	-	-	-	-
MedGemma-27B 2025.11	75.5	80.8	66.1	76.3	74	66.1	78.5	55.8	70.3	-	-	-	-	-	-
Med-R1 2026.04	73.5	-	-	82.2	-	-	83.7	45.2	71.1	68.8	-	-	68.9	77.4	69.5
HealthGPT-14B 2025.11	70.3	82.5	83.6	88	69	72.7	83.7	56.7	75.8	-	-	-	-	-	-
HuatuoGPT-Vision-7B 2025.11	70	84.2	72.1	85.7	72.5	75.7	81.9	81.2	77.9	-	-	-	-	-	-
Qwen3-VL 2026.01	68	-	-	85	83.67	81.67	86.33	66.33	79.33	79.33	84.33	79.33	-	-	-
Qwen2-VL-72B 2026.04	68	-	-	72.8	-	-	77.2	51.4	68.1	69.4	-	-	65.3	72.6	67.8
Qwen3-VL 2026.01	66.33	-	-	78.67	77	80.67	84	62.67	75.42	77.67	76.33	75.1	-	-	-
Qwen2.5-VL-72B 2026.04	66.2	-	-	69.2	-	-	77.6	49.8	67.7	68.7	-	-	69.8	71	69.4
Qwen3-VL 2026.01	65.67	-	-	63.33	67.33	82	81.33	58.67	69.42	70	67	69.33	-	-	-
ViTAR 2026.04	64	-	-	84.2	-	-	81	84.9	74.2	68.6	-	-	82.3	83.4	78.4
HuatuoGPT 2026.01	62.67	-	-	76.33	64	63.33	69.33	49.67	65.29	67.33	69.67	65	-	-	-
QWen2.5-VL-7B 2025.11	62.5	70.8	66.4	66.2	68	70.9	76.8	35.6	64.7	-	-	-	-	-	-
Qwen3-VL 2026.01	62	-	-	75.33	76.67	84	85.67	61.67	74.75	73.67	79	74.91	-	-	-
KEPO (τ = 1) 2026.01	62	-	-	82.33	78.67	81.67	85	74	80.25	96	82.33	78	-	-	-
Qwen2-VL-7B 2026.04	61.5	-	-	59.3	-	-	64.3	36	54.6	45.8	-	-	49.1	59.8	61.1
Qwen2.5-VL-7B 2026.04	60.4	-	-	61.2	-	-	74	30.7	60.3	58.4	-	-	62.5	67.3	67.8
BLIP-2† 2026.04	56.7	-	-	68.1	-	-	67.6	37.3	51	41.3	-	-	40.7	46.2	50.4
KEPO - KE (τ = 1) 2026.01	55.33	-	-	77.33	74.67	83.67	81	72	76.29	83.67	82.67	75.24	-	-	-
KEPO (τ = 0) 2026.01	54.67	-	-	78	79	83.67	82.67	72.33	78.71	95.67	83.67	76.29	-	-	-
Qwen2.5-VL-3B 2026.04	53.9	-	-	56.2	-	-	61.8	32.7	54.2	54.2	-	-	52.9	62.5	59.6
KEPO - KE (τ = 0) 2026.01	53.33	-	-	75.67	82	84.33	81.33	72.67	75.58	81.33	74	74.76	-	-	-
SFT Base 2026.01	53	-	-	67.67	64.67	78	79.67	55.33	69.79	85.33	74.67	67.57	-	-	-
On-policy KD (λ = 1) 2026.01	52.67	-	-	58.33	67	73.67	79.67	75	65.92	59.67	61.33	66.81	-	-	-
GRPO Base 2026.01	51	-	-	73.33	75	84	78	73	74.33	79	81.33	73.67	-	-	-
Supervised KD (λ = 0) 2026.01	50.67	-	-	73	73.33	83.33	78	74.67	73.92	78.67	79.67	73.24	-	-	-
Qwen3-VL 2026.01	50.33	-	-	55.33	68.33	75.67	79.67	74.33	67.46	64	72	67.95	-	-	-
Mixed KD (λ = 0.5) 2026.01	49.33	-	-	73	70.67	82.33	77.67	74.33	73.04	77.67	79.33	72.38	-	-	-
Qwen3-VL 2026.01	48.67	-	-	68.67	68	79.33	79	73.33	69.42	61.33	77	70.57	-	-	-
On-policy KD (λ = 1) 2026.01	48.67	-	-	73	71.33	82.67	78.67	74.67	73.13	78.33	77.67	72.38	-	-	-
KEPO - KE (τ = 0) 2026.01	48.33	-	-	73	70	80.33	77	71.67	71.33	77.67	79	72.13	-	-	-
GRPO Base 2026.01	48	-	-	73.67	71.67	80	78.33	72.33	73.12	80	81	72.14	-	-	-
Qwen2-VL-2B 2026.04	45.1	-	-	35.1	-	-	39.3	30.9	38.1	38.6	-	-	35.8	43.2	36.9
MedVInT† 2026.04	40.7	-	-	23.3	-	-	55.1	41.3	37.1	43.1	-	-	29.1	31.8	32
InternVL2 2026.04	40.2	-	-	59.1	-	-	57.9	49.1	54.2	58.1	-	-	51.9	53.2	64
LLaVA-Med-7B 2025.11	38.7	48.3	39.5	45.7	58.8	49.3	43.4	48.1	46.5	-	-	-	-	-	-
Med-Flamingo† 2026.04	38.5	-	-	26.5	-	-	30.3	24.6	30.4	40.6	-	-	32.4	30.1	19.9
HealthGPT 2026.04	35.5	-	-	89.3	-	-	81.9	51.4	68	78.5	-	-	64.9	54.6	88.2
HuatuoGPT-Vision 2026.04	35.3	-	-	59.3	-	-	41.5	60.1	50.4	40.4	-	-	53.1	51.4	62.3
LLaVA-NeXT-7B 2025.11	31.7	27	24.6	25.7	30.2	26.6	25.8	25.5	27.1	-	-	-	-	-	-
LLaVA-7B 2025.11	30.7	26.1	23.9	24.8	30.7	26.4	23	25.1	26	-	-	-	-	-	-
InstructBLIP† 2026.04	28.7	-	-	42.6	-	-	61	41.3	45.7	33.2	-	-	62.2	50.3	46.3
RadFM† 2026.04	27.6	-	-	32.8	-	-	31	16.6	29.5	24.1	-	-	39.2	36.9	28
MiniGPT-4† 2026.04	22.8	-	-	31.4	-	-	38.3	25.5	32.5	27.5	-	-	40.3	38.3	36.2
LLaMA Adapter v2† 2026.04	21.4	-	-	33	-	-	46.4	34.1	37.8	26.6	-	-	51.8	50.7	38.7
LLaVA-Med† 2026.04	18.7	-	-	34.6	-	-	30.7	29.9	32.3	27.5	-	-	45	39	33.3
LLaVA† 2026.04	17.7	-	-	33.7	-	-	30.7	18.7	31.7	26.7	-	-	49.7	47.1	28.9

M3LLM-8B

2025.11

85.1

85.5

89.3

90.2

79.8

83.6

88.7

78.2

85.7

-

MedLVR (ours)

2026.04

80.4

-

80.2

-

77.5

51.8

78

82

-

76.6

81.9

70.2

InternVL3-8B

2025.11

78.9

87.5

80.4

77.2

81.8

82.2

87.3

76.4

79

-

Lingshu-7B

2025.11

77.2

88.9

83.3

89.9

83.2

82.7

86.4

80.6

82.8

-

MedGemma-27B

2025.11

75.5

80.8

66.1

76.3

74

66.1

78.5

55.8

70.3

-

Med-R1

2026.04

73.5

-

82.2

-

83.7

45.2

71.1

68.8

-

68.9

77.4

69.5

HealthGPT-14B

2025.11

70.3

82.5

83.6

88

69

72.7

83.7

56.7

75.8

-

HuatuoGPT-Vision-7B

2025.11

70

84.2

72.1

85.7

72.5

75.7

81.9

81.2

77.9

-

Qwen3-VL

2026.01

68

-

85

83.67

81.67

86.33

66.33

79.33

84.33

79.33

-

Qwen2-VL-72B

2026.04

68

-

72.8

-

77.2

51.4

68.1

69.4

-

65.3

72.6

67.8

Qwen3-VL

2026.01

66.33

-

78.67

77

80.67

84

62.67

75.42

77.67

76.33

75.1

-

Qwen2.5-VL-72B

2026.04

66.2

-

69.2

-

77.6

49.8

67.7

68.7

-

69.8

71

69.4

Qwen3-VL

2026.01

65.67

-

63.33

67.33

82

81.33

58.67

69.42

70

67

69.33

-

ViTAR

2026.04

64

-

84.2

-

81

84.9

74.2

68.6

-

82.3

83.4

78.4

HuatuoGPT

2026.01

62.67

-

76.33

64

63.33

69.33

49.67

65.29

67.33

69.67

65

-

QWen2.5-VL-7B

2025.11

62.5

70.8

66.4

66.2

68

70.9

76.8

35.6

64.7

-

Qwen3-VL

2026.01

62

-

75.33

76.67

84

85.67

61.67

74.75

73.67

79

74.91

-

KEPO (τ = 1)

2026.01

62

-

82.33

78.67

81.67

85

74

80.25

96

82.33

78

-

Qwen2-VL-7B

2026.04

61.5

-

59.3

-

64.3

36

54.6

45.8

-

49.1

59.8

61.1

Qwen2.5-VL-7B

2026.04

60.4

-

61.2

-

74

30.7

60.3

58.4

-

62.5

67.3

67.8

BLIP-2†

2026.04

56.7

-

68.1

-

67.6

37.3

51

41.3

-

40.7

46.2

50.4

KEPO - KE (τ = 1)

2026.01

55.33

-

77.33

74.67

83.67

81

72

76.29

83.67

82.67

75.24

-

KEPO (τ = 0)

2026.01

54.67

-

78

79

83.67

82.67

72.33

78.71

95.67

83.67

76.29

-

Qwen2.5-VL-3B

2026.04

53.9

-

56.2

-

61.8

32.7

54.2

-

52.9

62.5

59.6

KEPO - KE (τ = 0)

2026.01

53.33

-

75.67

82

84.33

81.33

72.67

75.58

81.33

74

74.76

-

SFT Base

2026.01

53

-

67.67

64.67

78

79.67

55.33

69.79

85.33

74.67

67.57

-

On-policy KD (λ = 1)

2026.01

52.67

-

58.33

67

73.67

79.67

75

65.92

59.67

61.33

66.81

-

GRPO Base

2026.01

51

-

73.33

75

84

78

73

74.33

79

81.33

73.67

-

Supervised KD (λ = 0)

2026.01

50.67

-

73

73.33

83.33

78

74.67

73.92

78.67

79.67

73.24

-

Qwen3-VL

2026.01

50.33

-

55.33

68.33

75.67

79.67

74.33

67.46

64

72

67.95

-

Mixed KD (λ = 0.5)

2026.01

49.33

-

73

70.67

82.33

77.67

74.33

73.04

77.67

79.33

72.38

-

Qwen3-VL

2026.01

48.67

-

68.67

68

79.33

79

73.33

69.42

61.33

77

70.57

-

On-policy KD (λ = 1)

2026.01

48.67

-

73

71.33

82.67

78.67

74.67

73.13

78.33

77.67

72.38

-

KEPO - KE (τ = 0)

2026.01

48.33

-

73

70

80.33

77

71.67

71.33

77.67

79

72.13

-

GRPO Base

2026.01

48

-

73.67

71.67

80

78.33

72.33

73.12

80

81

72.14

-

Qwen2-VL-2B

2026.04

45.1

-

35.1

-

39.3

30.9

38.1

38.6

-

35.8

43.2

36.9

MedVInT†

2026.04

40.7

-

23.3

-

55.1

41.3

37.1

43.1

-

29.1

31.8

32

InternVL2

2026.04

40.2

-

59.1

-

57.9

49.1

54.2

58.1

-

51.9

53.2

64

LLaVA-Med-7B

2025.11

38.7

48.3

39.5

45.7

58.8

49.3

43.4

48.1

46.5

-

Med-Flamingo†

2026.04

38.5

-

26.5

-

30.3

24.6

30.4

40.6

-

32.4

30.1

19.9

HealthGPT

2026.04

35.5

-

89.3

-

81.9

51.4

68

78.5

-

64.9

54.6

88.2

HuatuoGPT-Vision

2026.04

35.3

-

59.3

-

41.5

60.1

50.4

40.4

-

53.1

51.4

62.3

LLaVA-NeXT-7B

2025.11

31.7

27

24.6

25.7

30.2

26.6

25.8

25.5

27.1

-

LLaVA-7B

2025.11

30.7

26.1

23.9

24.8

30.7

26.4

23

25.1

26

-

InstructBLIP†

2026.04

28.7

-

42.6

-

61

41.3

45.7

33.2

-

62.2

50.3

46.3

RadFM†

2026.04

27.6

-

32.8

-

31

16.6

29.5

24.1

-

39.2

36.9

28

MiniGPT-4†

2026.04

22.8

-

31.4

-

38.3

25.5

32.5

27.5

-

40.3

38.3

36.2

LLaMA Adapter v2†

2026.04

21.4

-

33

-

46.4

34.1

37.8

26.6

-

51.8

50.7

38.7

LLaVA-Med†

2026.04

18.7

-

34.6

-

30.7

29.9

32.3

27.5

-

45

39

33.3

LLaVA†

2026.04

17.7

-

33.7

-

30.7

18.7

31.7

26.7

-

49.7

47.1

28.9

Medical Visual Question Answering on OmniMedVQA (test)

Evaluation Results