Skip to content

Benchmark

IFEval

Instruction following benchmarks measure how closely models follow instructions. Scores are percentages. Higher is better.

Models measured
209
Scores recorded
219
Sources
3
IQ weight
0.84

Score distribution

57.9%
Mean
59.7%
Median
26.2%
Std dev
20.6%
10th percentile
89.8%
90th percentile
0%
Min
0%98.9%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.84 for chat models, measured over 209 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
151granite-3-1-8b-base· Hugging Face Open LLM Leaderboard data42.2%
152DeepSeek R1 Distill Qwen 32B· Hugging Face Open LLM Leaderboard data41.9%
153Qwen2.5 72B· Hugging Face Open LLM Leaderboard data41.4%
154Qwen2.5 32B· Hugging Face Open LLM Leaderboard data40.8%
155DeepSeek R1 Distill Qwen 7B· Hugging Face Open LLM Leaderboard data40.4%
156QwQ 32B Preview· Hugging Face Open LLM Leaderboard data40.4%
157Qwen2.5 Math 72B Instruct· Hugging Face Open LLM Leaderboard data40%
158Llama-2-7b-chat-hf· Hugging Face Open LLM Leaderboard data39.9%
159Llama-2-13b-chat-hf· Hugging Face Open LLM Leaderboard data39.8%
160Gemma 7B IT· Hugging Face Open LLM Leaderboard data38.7%
161Qwen 2 (72B)· Hugging Face Open LLM Leaderboard data38.2%
162DeepSeek R1 Distill Llama 8B· Hugging Face Open LLM Leaderboard data37.8%
163Qwen2.5 14B· Hugging Face Open LLM Leaderboard data36.9%
164qwen2p5-coder-14b· Hugging Face Open LLM Leaderboard data34.7%
165DeepSeek R1 Distill Qwen 1.5B· Hugging Face Open LLM Leaderboard data34.6%
166Qwen2.5-Coder-7B· Hugging Face Open LLM Leaderboard data34.5%
167Qwen2.5 7B· Hugging Face Open LLM Leaderboard data33.7%
168Qwen2 1.5b Instruct· Hugging Face Open LLM Leaderboard data33.7%
169Qwen1.5-4B-Chat· Hugging Face Open LLM Leaderboard data31.6%
170Qwen 2 (7B)· Hugging Face Open LLM Leaderboard data31.5%
171Qwen2.5 0.5b Instruct· Hugging Face Open LLM Leaderboard data30.7%
172Gemma 1.1 2B IT· Hugging Face Open LLM Leaderboard data30.7%
173Yi-34B· Hugging Face Open LLM Leaderboard data30.5%
174recurrentgemma-2b· Hugging Face Open LLM Leaderboard data30.2%
175Qwen1.5-14B· Hugging Face Open LLM Leaderboard data29.1%
176Yi-6B· Hugging Face Open LLM Leaderboard data28.9%
177Starcoder2 15b· Hugging Face Open LLM Leaderboard data27.8%
178phi-2· Hugging Face Open LLM Leaderboard data27.4%
179Gemma 2B IT· Hugging Face Open LLM Leaderboard data26.9%
180Qwen1.5-7B· Hugging Face Open LLM Leaderboard data26.8%
181Qwen2.5 1.5B· Hugging Face Open LLM Leaderboard data26.7%
182gemma-7b· Hugging Face Open LLM Leaderboard data26.6%
183Qwen2.5 Math 7B Instruct· Hugging Face Open LLM Leaderboard data26.4%
184mixtral-8x22b-v0.1· Hugging Face Open LLM Leaderboard data25.8%
185gemma2-27b· Hugging Face Open LLM Leaderboard data24.8%
186Qwen1.5-4B· Hugging Face Open LLM Leaderboard data24.4%
187Mixtral 8X7b V0.1· Hugging Face Open LLM Leaderboard data24.2%
188Mistral 7B v0.1· Hugging Face Open LLM Leaderboard data23.9%
189vicuna-7b-v1.5· Hugging Face Open LLM Leaderboard data23.5%
190RedPajama-INCITE-Base-3B-v1· Hugging Face Open LLM Leaderboard data22.9%
191mistral-7b-v0.3· Hugging Face Open LLM Leaderboard data22.7%
192qwen2-0.5b-instruct· Hugging Face Open LLM Leaderboard data22.5%
193Qwen1.5-1.8B· Hugging Face Open LLM Leaderboard data21.5%
194Llama-2-7B-32K-Instruct· Hugging Face Open LLM Leaderboard data21.3%
195RedPajama-INCITE-Instruct-3B-v1· Hugging Face Open LLM Leaderboard data21.2%
196Qwen 2 (1.5B)· Hugging Face Open LLM Leaderboard data21.1%
197RedPajama-INCITE-7B-Base· Hugging Face Open LLM Leaderboard data20.8%
198RedPajama-INCITE-7B-Instruct· Hugging Face Open LLM Leaderboard data20.6%
199gemma-2-9b· Hugging Face Open LLM Leaderboard data20.4%
200gemma-2b· Hugging Face Open LLM Leaderboard data20.4%

Showing 151–200 of 219 results.

Data credits: Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.