Skip to content

Benchmark

WildBench: haiku_reward.Creative Tasks.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
1.35

Score distribution

13.2
Mean
17.2
Median
25.9
Std dev
-16.8
10th percentile
44.5
90th percentile
-62.3
Min
-62.347.1
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.35 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-4 Turboallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data47.1
2GPT-4-0125-previewallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data45.8
3GPT 4oallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data45.4
4Gemini 1.5 Proallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data44.5
5Claude 3.5 Sonnetallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data40.1
6Reka Coreallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data35.4
7Llama 3 70B Instructallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data34.5
8Yi 1.5 34B Chatallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data33.6
9Gemini 1.5 Flashallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data32.2
10Nemotron-4 340B Instructallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data32.1
11yi-largeallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data31.9
12Claude 3 Opusallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data31.8
13DeepSeek Coder V2 Instructallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data28.3
14Command R Plusallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data27.2
15GLM-4-9b-chatallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data21.2
16Qwen2 72B Instructallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data17.2
17Claude 3 Sonnetallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data14.6
18Starling LM 7B Betaallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data12.4
19Mistral Largeallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307, 2402· WildBench direct data10.2
20Command R+allenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data7
21Llama 3 8B Instructallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data4.1
22Reka Flashallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data1.8
23Mixtral 8x7B Instruct v0.1allenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-2.3
24Mistral 7B Instruct v0.2allenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-3.2
25Tulu 2 DPO 70Ballenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-3.3
26Reka Edgeallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-3.6
27Llama-2-70b-chat-hfallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-6.8
28Nous Hermes 2 Mixtral 8x7B DPOallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-9.9
29GPT-3.5 Turboallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-16.8
30Llama-2-7b-chat-hfallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-25
31Gemma 7B ITallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-41.6
32Gemma 2B ITallenai/WildBench, haiku_reward.Creative Tasks.K: -1, claude-3-haiku-20240307· WildBench direct data-62.3

Data credits: WildBench, Allen Institute for AI