EVL

IFEval

Instruction following

benchmarkLanguage & Instruction% of verifiable instructions obeyed541 items

What this measures

Does the model actually *follow instructions*? Each prompt has checkable rules ('reply in exactly 3 bullet points,' 'don't use the letter e') and a program verifies compliance.

Sample

'Write a 2-paragraph summary of photosynthesis in all lowercase, with no commas.' — a script checks every constraint.

Leaderboard · 97 models

% of verifiable instructions obeyed — higher is better

1Llama-3.3-70B-Instructcurated · 2026-05

92.1

2Llama-3.1-70B-Instructcurated · 2026-05

87.5

3calme-2.1-qwen2.5-72bOpen LLM LB · 2026-05

86.6

4Llama-3.1-SauerkrautLM-70b-InstructOpen LLM LB · 2026-05

86.6

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

86.4

6DeepSeek-V3curated · 2026-05

86.1

7calme-2.3-llama3.1-70bOpen LLM LB · 2026-05

86.0

8calme-2.2-llama3.1-70bOpen LLM LB · 2026-05

85.9

9Qwen2.5-72B-Instruct-abliteratedOpen LLM LB · 2026-05

85.9

10ZYH-LLM-Qwen2.5-14B-V3Open LLM LB · 2026-05

85.8

11orca_mini_v8_1_70bOpen LLM LB · 2026-05

85.7

12RYS-Llama3.1-LargeOpen LLM LB · 2026-05

84.9

13Gilgamesh-72BOpen LLM LB · 2026-05

84.9

14calme-2.2-qwen2.5-72bOpen LLM LB · 2026-05

84.8

15test-2.5-72BOpen LLM LB · 2026-05

84.4

16calme-2.1-llama3.1-70bOpen LLM LB · 2026-05

84.3

17Rombos-LLM-V2.6-Qwen-14bOpen LLM LB · 2026-05

84.3

18Qwen2.5-95B-InstructOpen LLM LB · 2026-05

84.3

19virtuoso-small-v2-tensopolis-v1Open LLM LB · 2026-05

84.2

20solar-pro-preview-instructOpen LLM LB · 2026-05

84.2

21Qwen2.5-14B-Instruct-1M-GRPO-ReasoningOpen LLM LB · 2026-05

84.1

22Qwen2.5-14B-Instruct-1MOpen LLM LB · 2026-05

84.1

23Mistral-Large-Instruct-2411Open LLM LB · 2026-05

84.0

24Qwen2.5-14B-1M-YOYO-V3Open LLM LB · 2026-05

84.0

25Qwen2.5-14B-YOYO-V4Open LLM LB · 2026-05

84.0

26EXAONE-3.5-32B-InstructOpen LLM LB · 2026-05

83.9

27orca_mini_v9_2_70bOpen LLM LB · 2026-05

83.8

28Llama-3.1-Tulu-3-70BOpen LLM LB · 2026-05

83.8

29ZYH-LLM-Qwen2.5-14B-V4Open LLM LB · 2026-05

83.6

30Qwen2.5-32B-InstructOpen LLM LB · 2026-05

83.5

31Cheng-2Open LLM LB · 2026-05

83.4

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

83.3

33zetasepic-abliteratedV2-Qwen2.5-32B-Inst-BaseMerge-TIESOpen LLM LB · 2026-05

83.3

34Qwen2.5-14B-Instruct-abliterated-v2Open LLM LB · 2026-05

83.3

35Configurable-Llama-3.1-8B-InstructOpen LLM LB · 2026-05

83.1

36Q2.5-Veltha-14BOpen LLM LB · 2026-05

82.9

37Josiefied-Qwen2.5-14B-Instruct-abliterated-v4Open LLM LB · 2026-05

82.9

38Llama-3.1-Tulu-3-70B-DPOOpen LLM LB · 2026-05

82.8

39Virtuoso-Small-v2Open LLM LB · 2026-05

82.7

40Cheng-2-v1.1Open LLM LB · 2026-05

82.7

41Llama-3.1-Tulu-3-8BOpen LLM LB · 2026-05

82.7

42Dearly_Beloved-8B-TIESOpen LLM LB · 2026-05

82.7

43Awqward2.5-32B-InstructOpen LLM LB · 2026-05

82.5

44Linkbricks-Horizon-AI-Avengers-V3-32BOpen LLM LB · 2026-05

82.5

45Dinobot-Opus-14B-ExpOpen LLM LB · 2026-05

82.4

46Josiefied-abliteratedV4-Qwen2.5-14B-Inst-BaseMerge-TIESOpen LLM LB · 2026-05

82.4

47miscii-14b-1028Open LLM LB · 2026-05

82.4

48lambda-qwen2.5-14b-dpo-testOpen LLM LB · 2026-05

82.3

49llama3.1-kartoffeldes-70BOpen LLM LB · 2026-05

82.3

50Qwen2.5-14B-it-restoreOpen LLM LB · 2026-05

82.1

51Linkbricks-Horizon-AI-Avengers-V6-32BOpen LLM LB · 2026-05

82.1

52calme-2.2-llama3-70bOpen LLM LB · 2026-05

82.1

53huihui-ai-abliterated-Qwen2.5-32B-Inst-BaseMerge-TIESOpen LLM LB · 2026-05

82.1

54Qwen2.5-14B-YOYO-V4-p1Open LLM LB · 2026-05

82.0

55Qwen2.5-14B-Gutenberg-Instruct-SlerpenoOpen LLM LB · 2026-05

82.0

56Tulu-3.1-8B-SuperNovaOpen LLM LB · 2026-05

81.9

57huihui-ai-abliteratedV2-Qwen2.5-14B-Inst-BaseMerge-TIESOpen LLM LB · 2026-05

81.8

58CalmeRys-78B-Orpo-v0.1Open LLM LB · 2026-05

81.6

59calme-2.1-qwen2-72bOpen LLM LB · 2026-05

81.6

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

81.6

61shuttle-3Open LLM LB · 2026-05

81.5

62Llama_3.1_8b_Smarteaz_V1.01Open LLM LB · 2026-05

81.5

63Linkbricks-Horizon-AI-Korean-Avengers-V2-27BOpen LLM LB · 2026-05

81.5

64Qwen2.5-14B-InstructOpen LLM LB · 2026-05

81.4

65Linkbricks-Horizon-AI-Korean-Avengers-V3-27BOpen LLM LB · 2026-05

81.4

66EXAONE-3.5-7.8B-InstructOpen LLM LB · 2026-05

81.4

67calme-2.1-rys-78bOpen LLM LB · 2026-05

81.4

68calme-3.1-instruct-78bOpen LLM LB · 2026-05

81.4

69Llama3.1-IgneousIguana-8BOpen LLM LB · 2026-05

81.3

70li-14b-v0.4Open LLM LB · 2026-05

81.3

71openbuddy-llama3.3-70b-v24.1-131kOpen LLM LB · 2026-05

81.2

72Qwen2.5-THREADRIPPER-Medium-CensoredOpen LLM LB · 2026-05

81.1

73ZEUS-8B-V11Open LLM LB · 2026-05

81.0

74Virtuoso-LiteOpen LLM LB · 2026-05

81.0

75Meta-Llama-3-70B-InstructOpen LLM LB · 2026-05

81.0

76ichor_1.1-8B-Model_StockOpen LLM LB · 2026-05

81.0

77Llama-3.1-SuperNova-8B-Lite_TIES_with_BaseOpen LLM LB · 2026-05

81.0

78Qwen2.5-Gutenberg-Doppel-14BOpen LLM LB · 2026-05

80.9

79FluentlyLM-PrinumOpen LLM LB · 2026-05

80.9

80Llama3-TenyxChat-70BOpen LLM LB · 2026-05

80.9

81tempmotacilla-cinerea-0308Open LLM LB · 2026-05

80.8

82lambda-qwen2.5-32b-dpo-testOpen LLM LB · 2026-05

80.8

83Llama-3.1-8B-Ultra-InstructOpen LLM LB · 2026-05

80.8

84Llama_3.1_8b_Typhoon_v1.03Open LLM LB · 2026-05

80.8

85Meta-LLama-3-Cat-Smaug-LLama-70bOpen LLM LB · 2026-05

80.7

86virtuoso-lite-tensopolis-v1Open LLM LB · 2026-05

80.7

87calme-2.3-rys-78bOpen LLM LB · 2026-05

80.7

88Qwen2.5-Lumen-14BOpen LLM LB · 2026-05

80.6

89calme-3.2-instruct-78bOpen LLM LB · 2026-05

80.6

90tempesthenno-kto-0205-ckpt80Open LLM LB · 2026-05

80.5

91Llama_3.1_8b_DodoWild_v2.10Open LLM LB · 2026-05

80.5

92MT4-Gen2-gemma-2-9BOpen LLM LB · 2026-05

80.5

93RYS-Llama-3-Large-InstructOpen LLM LB · 2026-05

80.5

94Llama-3.1-Tulu-3-70B-SFTOpen LLM LB · 2026-05

80.5

95MT5-gemma-2-9BOpen LLM LB · 2026-05

80.5

96Qwen2.5-14B-YOYO-V4-p2Open LLM LB · 2026-05

80.5

97BBAIIFEV1Open LLM LB · 2026-05

80.5

Benchmark health

active

Discriminating but climbing — the frontier is moving up this benchmark.

86.9

top-10 mean

92.1

best

models

What improves this score

Datasets & environments shown to raise it.

—+10.0

Tülu 3 SFT Mixture+9.0

UltraFeedback+6.0

—+6.0

LMSYS-Chat-1M+3.0

WebArena+1.5

Research using this benchmark

all →

VibeThinker-3B: Exploring the Frontier of Verifiable Reasoning in Small Language Models Instruction-Following Evaluation for Large Language Models Recovering Diversity Without Losing Alignment: A DPO Recipe for Post-Trained LLMs Latent Performance Profiling of Large Language Models Soft-SVeRL: Self-Verified Reinforcement Learning with Soft Rewards

Matched by name in title/abstract.