Evals/MATH Lvl 5

EVL

MATH Lvl 5

Hardest competition math

benchmarkMathAccuracy on Level-5 problems only1.3K items

What this measures

Only the *hardest* tier (Level 5) of the MATH competition set — the problems most models still fail. A frontier test of mathematical reasoning.

Sample

A Level-5 olympiad problem requiring several non-obvious steps and a clever insight to solve.

Leaderboard · 100 models

Accuracy on Level-5 problems only — higher is better

1AceMath-72B-InstructOpen LLM LB · 2026-05

71.5

2AceMath-7B-InstructOpen LLM LB · 2026-05

63.4

3AceInstruct-72BOpen LLM LB · 2026-05

62.6

4Qwen2.5-32B-InstructOpen LLM LB · 2026-05

62.5

5Qwen2.5-Math-72B-InstructOpen LLM LB · 2026-05

62.4

6Awqward2.5-32B-InstructOpen LLM LB · 2026-05

62.3

7Linkbricks-Horizon-AI-Avengers-V6-32BOpen LLM LB · 2026-05

62.2

8Linkbricks-Horizon-AI-Avengers-V3-32BOpen LLM LB · 2026-05

61.8

9lambda-qwen2.5-32b-dpo-testOpen LLM LB · 2026-05

61.0

10Linkbricks-Horizon-AI-Avengers-V1-32BOpen LLM LB · 2026-05

60.3

11Qwen2.5-72B-Instruct-abliteratedOpen LLM LB · 2026-05

60.1

12WomboCombo-R1-Coder-14B-PreviewOpen LLM LB · 2026-05

59.9

13Qwen2.5-72B-InstructOpen LLM LB · 2026-05

59.8

14qwen2.5-test-32b-itOpen LLM LB · 2026-05

59.7

15Qwen2.5-32B-Instruct-abliterated-v2Open LLM LB · 2026-05

59.5

16huihui-ai-abliterated-Qwen2.5-32B-Inst-BaseMerge-TIESOpen LLM LB · 2026-05

59.4

17calme-2.1-qwen2.5-72bOpen LLM LB · 2026-05

59.1

18calme-2.2-qwen2.5-72bOpen LLM LB · 2026-05

58.9

19zetasepic-abliteratedV2-Qwen2.5-32B-Inst-BaseMerge-TIESOpen LLM LB · 2026-05

58.5

20Qwen2.5-72B-2x-Instruct-TIES-v1.0Open LLM LB · 2026-05

57.9

21Lion-Lamarck-v.1.1.0Open LLM LB · 2026-05

57.6

22Gauss-Opus-14B-R999Open LLM LB · 2026-05

57.6

23DeepSeek-R1-Distill-Qwen-14BOpen LLM LB · 2026-05

57.0

24Sombrero-Opus-14B-Sm1Open LLM LB · 2026-05

56.6

25Linkbricks-Horizon-AI-Avengers-V2-32BOpen LLM LB · 2026-05

56.6

26Viper-Coder-v1.6-r999Open LLM LB · 2026-05

56.6

27o-distil-qwenOpen LLM LB · 2026-05

56.5

28Lion-Lamarck-v.1.0.9Open LLM LB · 2026-05

56.4

29Chocolatine-14B-Instruct-DPO-v1.3Open LLM LB · 2026-05

56.2

30Apollo-70BOpen LLM LB · 2026-05

56.1

31li-14b-v0.4Open LLM LB · 2026-05

55.7

32NQLSG-Qwen2.5-14B-MegaFusion-v8Open LLM LB · 2026-05

55.6

33DS-R1-Distill-Q2.5-14B-Harmony_V0.1Open LLM LB · 2026-05

55.5

34tempmotacilla-cinerea-0308Open LLM LB · 2026-05

55.5

35Lion-Lamarck-v.1.0.8Open LLM LB · 2026-05

55.4

36DeepSeek-R1-Distill-Qwen-25.5B-BrainstormOpen LLM LB · 2026-05

55.4

37Qwen2-Math-72B-InstructOpen LLM LB · 2026-05

55.4

38Qwen2.5-14B-InstructOpen LLM LB · 2026-05

55.3

39Coma-II-14BOpen LLM LB · 2026-05

55.1

40Qwen2.5-14B-InstructOpen LLM LB · 2026-05

54.8

41huihui-ai-abliteratedV2-Qwen2.5-14B-Inst-BaseMerge-TIESOpen LLM LB · 2026-05

54.8

42BigQwen2.5-52B-InstructOpen LLM LB · 2026-05

54.8

43NQLSG-Qwen2.5-14B-MegaFusion-v9.1Open LLM LB · 2026-05

54.7

44Viper-Coder-v1.1Open LLM LB · 2026-05

54.6

45lambda-qwen2.5-14b-dpo-testOpen LLM LB · 2026-05

54.6

46Linkbricks-Horizon-AI-Avengers-V5-32BOpen LLM LB · 2026-05

54.6

47QWEN-14B-B100Open LLM LB · 2026-05

54.4

48Cheng-2Open LLM LB · 2026-05

54.4

49Josiefied-Qwen2.5-14B-Instruct-abliterated-v4Open LLM LB · 2026-05

54.2

50sky-t1-coder-32b-flashOpen LLM LB · 2026-05

54.2

51Rombos-LLM-V2.5-Qwen-72bOpen LLM LB · 2026-05

54.2

52Qwen2.5-Gutenberg-Doppel-14BOpen LLM LB · 2026-05

54.2

53NQLSG-Qwen2.5-14B-MegaFusion-v8.7Open LLM LB · 2026-05

54.1

54Lamarckvergence-14BOpen LLM LB · 2026-05

54.0

55FluentlyLM-PrinumOpen LLM LB · 2026-05

54.0

56Dyson-14bOpen LLM LB · 2026-05

53.9

57Cheng-2-v1.1Open LLM LB · 2026-05

53.9

58ZYH-LLM-Qwen2.5-14B-V4Open LLM LB · 2026-05

53.9

59RomboUltima-32BOpen LLM LB · 2026-05

53.9

60Qwen2.5-14B-it-restoreOpen LLM LB · 2026-05

53.7

61NQLSG-Qwen2.5-14B-MegaFusion-v8.9Open LLM LB · 2026-05

53.7

62Qwen2.5-Lumen-14BOpen LLM LB · 2026-05

53.6

63Linkbricks-Horizon-AI-Avengers-V4-32BOpen LLM LB · 2026-05

53.6

64Sombrero-Opus-14B-Elite5Open LLM LB · 2026-05

53.5

65Qwen2.5-14B-1M-YOYO-V3Open LLM LB · 2026-05

53.5

66ultiima-72BOpen LLM LB · 2026-05

53.5

67Qwen2.5-14B-YOYO-V4Open LLM LB · 2026-05

53.5

68Qwen2.5-THREADRIPPER-Medium-CensoredOpen LLM LB · 2026-05

53.4

69Qwen2.5-14B-YOYO-V4-p1Open LLM LB · 2026-05

53.3

70li-14b-v0.4-slerp0.1Open LLM LB · 2026-05

53.3

71NQLSG-Qwen2.5-14B-MegaFusion-v9.2Open LLM LB · 2026-05

53.3

72Flashlight-v1.1Open LLM LB · 2026-05

53.2

73Qwen2.5-14B-Gutenberg-Instruct-SlerpenoOpen LLM LB · 2026-05

53.2

74Imbue-14bOpen LLM LB · 2026-05

53.2

75QwQ-32B-Coder-Fusion-9010Open LLM LB · 2026-05

53.2

76Dinobot-Opus-14B-ExpOpen LLM LB · 2026-05

53.2

77Josiefied-abliteratedV4-Qwen2.5-14B-Inst-BaseMerge-TIESOpen LLM LB · 2026-05

53.2

78Qwen2.5-14B-Instruct-1M-GRPO-ReasoningOpen LLM LB · 2026-05

53.0

79Qwen2.5-14B-Instruct-1MOpen LLM LB · 2026-05

53.0

80Qwen2.5-14B-Instruct-abliterated-v2Open LLM LB · 2026-05

53.0

81Qwen2.5-95B-InstructOpen LLM LB · 2026-05

53.0

82AceInstruct-7BOpen LLM LB · 2026-05

52.9

83qwen2.5-14b-tensopolis-v1Open LLM LB · 2026-05

52.9

84Lix-14B-v0.1Open LLM LB · 2026-05

52.9

85Qwen2.5-14B-HyperMarck-dlOpen LLM LB · 2026-05

52.9

86ZYH-LLM-Qwen2.5-14B-V3Open LLM LB · 2026-05

52.7

87mocha-14BOpen LLM LB · 2026-05

52.6

88Qwen2.5-14B-Gutenberg-1e-DeltaOpen LLM LB · 2026-05

52.6

89MFDOOM-14BOpen LLM LB · 2026-05

52.6

90Blossom-V6-14BOpen LLM LB · 2026-05

52.6

91CogitoZOpen LLM LB · 2026-05

52.4

92SZA-14B-sceOpen LLM LB · 2026-05

52.4

93Qwen2.5-72B-Instruct-abliteratedOpen LLM LB · 2026-05

52.4

94Condor-Opus-14B-ExpOpen LLM LB · 2026-05

52.3

95Qwentile2.5-32B-InstructOpen LLM LB · 2026-05

52.2

96Rombos-LLM-V2.6-Qwen-14bOpen LLM LB · 2026-05

52.1

97Alien-CoT-14B-sceOpen LLM LB · 2026-05

52.0

98Qwen2.5-14B-ReasoningMergeOpen LLM LB · 2026-05

52.0

99Lo-Phi-14bOpen LLM LB · 2026-05

52.0

100RZA-14B-sceOpen LLM LB · 2026-05

51.9

Benchmark health

headroom

Plenty of room at the top — this benchmark still separates frontier models clearly.

63.0

top-10 mean

71.5

best

100

models

What improves this score

Datasets & environments shown to raise it.

No data yet.