IFEval leaderboard

A benchmark of 541 prompts built from 25 verifiable instruction types. It tests whether a model follows checkable constraints such as keyword, length, casing, and response-format requirements.

Models ranked by their published Instruction-Following Eval value

Rank	Model	Provider	Constrained generation
1	Qwen3.5-27B	Alibaba	95
2	Agents-A1	InternScience	94.82
3	Agents-A1-4B	InternScience	94.8
4	Qwen3.7 Plus	Alibaba	94.6
5	Qwen3.6 Plus	Alibaba	94.3
5	Qwen3.7 Max	Alibaba	94.3
7	DSdots3-note Preview	Dots Studio	93.9
7	Kimi K2.5	Moonshot AI	93.9
7	o3-mini	OpenAI	93.9
10	Qwen3.5-122B-A10B	Alibaba	93.4
11	GLM-5	Z.AI	92.6
11	Qwen3.5 397B	Alibaba	92.6
13	K-EXAONE 2.0	LG AI Research	92.4
14	o1	OpenAI	92.2
15	Qwen3.5-35B-A3B	Alibaba	91.9
16	LFM2.5-8B-A1B	LiquidAI	91.84
17	PMTernary Bonsai 2 27B	Prism ML	91.31
18	Claude Opus 4.5	Anthropic	90.9
19	GPT-4.1 mini	OpenAI	88.5
20	GPT-4.1	OpenAI	87.4
21	OPMiniCPM5-2B	OpenBMB	86.7
22	DeepSeek V3	DeepSeek	86.1
23	ZYZAYA1-8B	Zyphra	85.58
24	GPT-4.1 nano	OpenAI	83.2
25	LFM2.5-VL-3B	LiquidAI	82.3
26	KAKanana-2 3B Instruct	Kakao	80.96
27	CECeleris-1	Celeris	80.8
28	OPMiniCPM5-1B	OpenBMB	80.41
29	KAKanana-2 1.3B Instruct	Kakao	77.63
30	JEMellum2-12B-A2.5B-Thinking	JetBrains	76.5
31	JEMellum2-12B-A2.5B-Instruct	JetBrains	75.8
32	LFM2.5-230M	LiquidAI	71.71
33	LFM2.5-VL-450M	LiquidAI	61.16

Evidence key: Observed

Rows are ordered by the value Instruction-Following Evaluation for Large Language Models published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.