Instruction Following Benchmark

IFBench evaluates precise instruction-following generalization on 58 challenging, verifiable out-of-domain constraints. Unlike IFEval which tests familiar constraint types, IFBench specifically measures how well models follow novel instructions they haven't been optimized for, exposing overfitting to common instruction patterns.

In index
Instruction
Best in instruction

Models57

Top result89.2%Flint

Top-3 spread6.4 pts

Year2025

Result and price

24 of 57 results

Results

57 results

#ModelResult

01Flint89.2%

02MAI-Thinking-185.0%

03Qwen3.8 Max82.8%

04Inkling-Small82.2%

05Nemotron 3 Ultra81.7%

06Sage-A1.581.6%

07Qwen3.8-Omni-Flash81.5%

08Qwen3.8-Flash-Next81.3%

08Grok 4.381.3%

10Beam 1.0 501B-A23B FP880.7%

11dots3-note Preview80.4%

12Solar Open 280.0%

13Inkling79.8%

14Beam 1.0 501B-A23B79.6%

15Qwen3.8-27B79.5%

16Granite 4.2 8BAbove fit79.3%

17Beam 1.0 501B-A23B NVFP479.2%

17Ripple-65B Preview79.2%

19Qwen3.7 Max79.1%

19Qwen3.7 Plus79.1%

21Granite 4.2 30BAbove fit77.2%

22Muse Glimmer 30B77.0%

22Mercury 2.577.0%

24Mercury 376.8%

25Qwen3.5 397B76.5%

26Gemini 3.5 Flash76.3%

27A.X K275.9%

28Qwen3.6 Plus75.8%

28Spark-X2.6-4B75.8%

30LFM2.5-24B-A2BAbove fit74.7%

31Ling 3.0 Flash74.5%

32Granite 4.2 3BAbove fit74.3%

33Nemotron 3 Nano Omni 30B A3BAbove fit74.2%

34Ternary Bonsai 2 27B74.0%

35Ling 3.0 Flash FP873.4%

36LFM3-1.7B73.0%

37Nemotron 3.5 Lightning 30B A3B NVFP4Above fit72.9%

38K-EXAONE 2.072.6%

39Nemotron 3.5 Super VL 120B A12B NVFP471.5%

40Nemotron 3.5 Super VL 120B A12B70.9%

41Agents-A1-4B69.1%

42MiniCPM5-2B66.3%

43Hy3 Preview63.1%

44LFM2.5-2.6B59.2%

45Claude Opus 4.558.0%

46Ling 2.6 Flash57.0%

47LFM2.5-8B-A1B56.5%

48OrcaCyber-Nano-1.056.3%

49Solar Pro 355.8%

50ZAYA1-8B52.6%

51Trinity-Large-Thinking52.3%

52MiniCPM5-1BAbove fit46.7%

53LFM2.5-230M38.4%

54Kanana-2 1.3B Instruct34.7%

55Kanana-2 3B Instruct33.3%

56LFM2.5-VL-3B25.8%

57LLaDA2.2-mini24.9%