Skip to content

Benchmark

WildBench: WB_Elo

People compare model answers and choose which they prefer. Scores are Elo ratings. Higher ratings are better.

Models measured
41
Scores recorded
41
Sources
1
IQ weight
1.61

Score distribution

1,154.3
Mean
1,162.8
Median
53.8
Std dev
1,088.1
10th percentile
1,210
90th percentile
996.6
Min
996.61,239.5
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.61 for chat models, measured over 41 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, WB_Elo· WildBench direct data1,239.5
2Claude 3.5 Sonnetallenai/WildBench, WB_Elo· WildBench direct data1,226.3
3GPT-4 Turboallenai/WildBench, WB_Elo· WildBench direct data1,222.7
4Gemini 1.5 Proallenai/WildBench, WB_Elo· WildBench direct data1,221.1
5DeepSeek V2 Chatallenai/WildBench, WB_Elo· WildBench direct data1,210
6GPT-4-0125-previewallenai/WildBench, WB_Elo· WildBench direct data1,208.3
7Claude 3 Opusallenai/WildBench, WB_Elo· WildBench direct data1,207.5
8Athene 70Ballenai/WildBench, WB_Elo· WildBench direct data1,204.6
9Mistral Large 2allenai/WildBench, WB_Elo· WildBench direct data1,197.7
10Gemini 1.5 Flashallenai/WildBench, WB_Elo· WildBench direct data1,196.8
11Llama 3 70B Instructallenai/WildBench, WB_Elo· WildBench direct data1,194.6
12GPT-4o miniallenai/WildBench, WB_Elo· WildBench direct data1,191.7
13yi-largeallenai/WildBench, WB_Elo· WildBench direct data1,189.5
14Nemotron-4 340B Instructallenai/WildBench, WB_Elo· WildBench direct data1,184.4
15DeepSeek Coder V2 Instructallenai/WildBench, WB_Elo· WildBench direct data1,183.5
16Claude 3 Sonnetallenai/WildBench, WB_Elo· WildBench direct data1,180.9
17Gemma 2 9B IT SimPOallenai/WildBench, WB_Elo· WildBench direct data1,180.5
18Qwen2 72B Instructallenai/WildBench, WB_Elo· WildBench direct data1,176.4
19Reka Coreallenai/WildBench, WB_Elo· WildBench direct data1,174.7
20Mistral Nemoallenai/WildBench, WB_Elo· WildBench direct data1,166.8
21Yi 1.5 34B Chatallenai/WildBench, WB_Elo· WildBench direct data1,162.8
22Claude 3 Haikuallenai/WildBench, WB_Elo· WildBench direct data1,159.8
23Gemma 2 9B ITallenai/WildBench, WB_Elo· WildBench direct data1,159.3
24Mistral Largeallenai/WildBench, WB_Elo, 2402· WildBench direct data1,158
25Command R Plusallenai/WildBench, WB_Elo· WildBench direct data1,151.2
26GLM-4-9b-chatallenai/WildBench, WB_Elo· WildBench direct data1,146.7
27Llama 3 8B Instructallenai/WildBench, WB_Elo· WildBench direct data1,134.2
28Reka Flashallenai/WildBench, WB_Elo· WildBench direct data1,130.7
29GPT-3.5 Turboallenai/WildBench, WB_Elo· WildBench direct data1,124.5
30Mixtral 8x7B Instruct v0.1allenai/WildBench, WB_Elo· WildBench direct data1,122.8
31Command R+allenai/WildBench, WB_Elo· WildBench direct data1,119.8
32Starling LM 7B Betaallenai/WildBench, WB_Elo· WildBench direct data1,119.3
33Gemma 2 2B ITallenai/WildBench, WB_Elo· WildBench direct data1,115
34Tulu 2 DPO 70Ballenai/WildBench, WB_Elo· WildBench direct data1,114
35Reka Edgeallenai/WildBench, WB_Elo· WildBench direct data1,098.5
36Mistral 7B Instruct v0.2allenai/WildBench, WB_Elo· WildBench direct data1,095.8
37Llama-2-70b-chat-hfallenai/WildBench, WB_Elo· WildBench direct data1,088.1
38Nous Hermes 2 Mixtral 8x7B DPOallenai/WildBench, WB_Elo· WildBench direct data1,080.5
39Gemma 7B ITallenai/WildBench, WB_Elo· WildBench direct data1,058.2
40Llama-2-7b-chat-hfallenai/WildBench, WB_Elo· WildBench direct data1,033.4
41Gemma 2B ITallenai/WildBench, WB_Elo· WildBench direct data996.6

Data credits: WildBench, Allen Institute for AI