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

Models42

Top result85.0%MAI-Thinking-1

Top-3 spread2.8 pts

Year2025

Result and price

Results

42 results

#ModelResult

01

MAI-Thinking-185.0%

02

Qwen3.8 Max82.8%

03TMInkling-Small82.2%

04

Nemotron 3 Ultra81.7%

05

Qwen3.8-Omni-Flash81.5%

06

Qwen3.8-Flash-Next81.3%

06

Grok 4.381.3%

08DSdots3-note Preview80.4%

09UPSolar Open 280.0%

10TMInkling79.8%

11

Qwen3.8-27B79.5%

12IBGranite 4.2 8B79.3%

13

Qwen3.7 Max79.1%

13

Qwen3.7 Plus79.1%

15IBGranite 4.2 30B77.2%

16

Muse Glimmer 30B77.0%

16INMercury 2.577.0%

18

Gemini 3.5 Flash76.3%

19STA.X K275.9%

20

Qwen3.6 Plus75.8%

21INLing 3.0 Flash74.5%

22IBGranite 4.2 3B74.3%

23

Nemotron 3 Nano Omni 30B A3B74.2%

24PMTernary Bonsai 2 27B74.0%

25INLing 3.0 Flash FP873.4%

26

Nemotron 3.5 Lightning 30B A3B NVFP472.9%

27LAK-EXAONE 2.072.6%

28INAgents-A1-4B69.1%

29OPMiniCPM5-2B66.3%

30TEHy3 Preview63.1%

31LILFM2.5-2.6B59.2%

32

Claude Opus 4.558.0%

33INLing 2.6 Flash57.0%

34LILFM2.5-8B-A1B56.5%

35UPSolar Pro 355.8%

36ZYZAYA1-8B52.6%

37OPMiniCPM5-1B46.7%

38LILFM2.5-230M38.4%

39KAKanana-2 1.3B Instruct34.7%

40KAKanana-2 3B Instruct33.3%

41LILFM2.5-VL-3B25.8%

42INLLaDA2.2-mini24.9%