Skip to content

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.

42 models

Higher is better.

  1. 1MAI-Thinking-1Microsoft85.0
  2. 2Qwen3.8 MaxAlibaba82.8
  3. 3Inkling-SmallThinking Machines Lab82.2
  4. 4Nemotron 3 UltraNVIDIA81.7
  5. 5Qwen3.8-Omni-FlashAlibaba81.5
  6. 6Qwen3.8-Flash-NextAlibaba81.3
  7. 7Grok 4.3xAI81.3
  8. 8dots3-note PreviewDots Studio80.4
  9. 9Solar Open 2Upstage80.0
  10. 10InklingThinking Machines Lab79.8
  11. 11Beam 1.0 501B-A23BReflection AI79.6
  12. 12Qwen3.8-27BAlibaba79.5
  13. 13Granite 4.2 8BIBM79.3
  14. 14Qwen3.7 MaxAlibaba79.1
  15. 15Qwen3.7 PlusAlibaba79.1
  16. 16Muse Glimmer 30BMeta77.0
  17. 17Mercury 2.5Inception77.0
  18. 18Mercury 3Inception76.8
  19. 19Qwen3.5 397BAlibaba76.5
  20. 20Gemini 3.5 FlashGoogle76.3
  21. 21A.X K2SK Telecom75.9
  22. 22Qwen3.6 PlusAlibaba75.8
  23. 23LFM2.5-24B-A2BLiquid AI74.7
  24. 24Ling 3.0 FlashInclusionAI74.5
  25. 25Nemotron 3 Nano Omni 30B A3BNVIDIA74.2
  26. 26Ternary Bonsai 2 27BPrism ML74.0
  27. 27Nemotron 3.5 Lightning 30B A3B NVFP4NVIDIA72.9
  28. 28K-EXAONE 2.0LG AI Research72.6
  29. 29Nemotron 3.5 Super VL 120B A12BNVIDIA70.9
  30. 30MiniCPM5-2BOpenBMB66.3
  31. 31Hy3 PreviewTencent63.1
  32. 32LFM2.5-2.6BLiquid AI59.2
  33. 33Claude Opus 4.5Anthropic58.0
  34. 34Ling 2.6 FlashInclusionAI57.0
  35. 35LFM2.5-8B-A1BLiquid AI56.5
  36. 36Solar Pro 3Upstage55.8
  37. 37ZAYA1-8BZyphra52.6
  38. 38Trinity-Large-ThinkingArcee AI52.3
  39. 39MiniCPM5-1BOpenBMB46.7
  40. 40LFM2.5-230MLiquid AI38.4
Higher is better. Showing the top 40 of 42. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Tasks
58

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.

All results

1MAI-Thinking-1 Microsoft85.0
2Qwen3.8 Max Alibaba82.8
3Inkling-Small Thinking Machines Lab82.2
4Nemotron 3 Ultra NVIDIA81.7
5Qwen3.8-Omni-Flash Alibaba81.5
6Qwen3.8-Flash-Next Alibaba81.3
7Grok 4.3 xAI81.3
8dots3-note Preview Dots Studio80.4
9Solar Open 2 Upstage80.0
10Inkling Thinking Machines Lab79.8
11Beam 1.0 501B-A23B Reflection AI79.6
12Qwen3.8-27B Alibaba79.5
13Granite 4.2 8B IBM79.3
14Qwen3.7 Max Alibaba79.1
15Qwen3.7 Plus Alibaba79.1
16Muse Glimmer 30B Meta77.0
17Mercury 2.5 Inception77.0
18Mercury 3 Inception76.8
19Qwen3.5 397B Alibaba76.5
20Gemini 3.5 Flash Google76.3
21A.X K2 SK Telecom75.9
22Qwen3.6 Plus Alibaba75.8
23LFM2.5-24B-A2B Liquid AI74.7
24Ling 3.0 Flash InclusionAI74.5
25Nemotron 3 Nano Omni 30B A3B NVIDIA74.2
26Ternary Bonsai 2 27B Prism ML74.0
27Nemotron 3.5 Lightning 30B A3B NVFP4 NVIDIA72.9
28K-EXAONE 2.0 LG AI Research72.6
29Nemotron 3.5 Super VL 120B A12B NVIDIA70.9
30MiniCPM5-2B OpenBMB66.3
31Hy3 Preview Tencent63.1
32LFM2.5-2.6B Liquid AI59.2
33Claude Opus 4.5 Anthropic58.0
34Ling 2.6 Flash InclusionAI57.0
35LFM2.5-8B-A1B Liquid AI56.5
36Solar Pro 3 Upstage55.8
37ZAYA1-8B Zyphra52.6
38Trinity-Large-Thinking Arcee AI52.3
39MiniCPM5-1B OpenBMB46.7
40LFM2.5-230M Liquid AI38.4
41Kanana-2 3B Instruct Kakao33.3
42LFM2.5-VL-3B Liquid AI25.8