Skip to content

Benchmark

WildBench: haiku_reward.Information/Advice seeking.K=-1

People compare model answers and choose which they prefer. Higher scores are better.

Models measured
32
Scores recorded
32
Sources
1
IQ weight
1.37

Score distribution

5.6
Mean
13.1
Median
28.9
Std dev
-33.5
10th percentile
41.1
90th percentile
-75.4
Min
-75.446.2
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.37 for chat models, measured over 32 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1GPT 4oallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data46.2
2Claude 3.5 Sonnetallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data42.4
3GPT-4 Turboallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data41.5
4Gemini 1.5 Proallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data41.1
5GPT-4-0125-previewallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data38.4
6Llama 3 70B Instructallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data32
7Claude 3 Opusallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data29.7
8Yi 1.5 34B Chatallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data27.5
9Nemotron-4 340B Instructallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data27.3
10yi-largeallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data27
11Gemini 1.5 Flashallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data26.7
12Reka Coreallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data24.6
13DeepSeek Coder V2 Instructallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data16.5
14Claude 3 Sonnetallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data15.9
15Command R Plusallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data15.4
16Qwen2 72B Instructallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data13.1
17GLM-4-9b-chatallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data12.8
18Command R+allenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-0.131
19Starling LM 7B Betaallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-1.9
20Llama 3 8B Instructallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-2.1
21Reka Flashallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-3.6
22Mistral Largeallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307, 2402· WildBench direct data-8.4
23Llama-2-70b-chat-hfallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-9.4
24Mixtral 8x7B Instruct v0.1allenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-10.9
25Reka Edgeallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-13.4
26Mistral 7B Instruct v0.2allenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-15.4
27Tulu 2 DPO 70Ballenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-16.8
28Nous Hermes 2 Mixtral 8x7B DPOallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-19.6
29GPT-3.5 Turboallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-33.5
30Llama-2-7b-chat-hfallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-34.1
31Gemma 7B ITallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-54.7
32Gemma 2B ITallenai/WildBench, haiku_reward.Information/Advice seeking.K=-1, claude-3-haiku-20240307· WildBench direct data-75.4

Data credits: WildBench, Allen Institute for AI