Skip to content

Benchmark

WildBench: gpt4t_reward.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
1.39

Score distribution

-37.1
Mean
-34.1
Median
21.9
Std dev
-64.8
10th percentile
-4.4
90th percentile
-86.1
Min
-86.11.7
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.39 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, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data1.7
2Gemini 1.5 Proallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-1.3
3GPT-4-0125-previewallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-4.2
4Claude 3.5 Sonnetallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-4.4
5Gemini 1.5 Flashallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-11.3
6Yi 1.5 34B Chatallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-17.3
7Llama 3 70B Instructallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-18.4
8Nemotron-4 340B Instructallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-21
9Claude 3 Opusallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-21.2
10yi-largeallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-21.4
11Reka Coreallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-21.7
12DeepSeek Coder V2 Instructallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-26.4
13GLM-4-9b-chatallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-30.1
14Claude 3 Sonnetallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-30.3
15Command R Plusallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-33.2
16Qwen2 72B Instructallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-34.1
17Claude 3 Haikuallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-44.6
18Command R+allenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-44.8
19Llama 3 8B Instructallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-46.3
20Mistral Largeallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09, 2402· WildBench direct data-46.4
21Reka Flashallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-47
22Starling LM 7B Betaallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-47
23Mixtral 8x7B Instruct v0.1allenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-50.1
24Llama-2-70b-chat-hfallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-53.4
25Reka Edgeallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-54.3
26Mistral 7B Instruct v0.2allenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-54.7
27Nous Hermes 2 Mixtral 8x7B DPOallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-54.7
28Tulu 2 DPO 70Ballenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-54.9
29GPT-3.5 Turboallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-64.8
30Llama-2-7b-chat-hfallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-66.6
31Gemma 7B ITallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-77.2
32Gemma 2B ITallenai/WildBench, gpt4t_reward.K: -1, gpt-4-turbo-2024-04-09· WildBench direct data-86.1

Data credits: WildBench, Allen Institute for AI