Skip to content

Benchmark

WildBench: haiku_reward.Math & Data Analysis.K=-1

Math benchmarks measure how well models solve maths problems. Higher scores are better.

Models measured
32
Scores recorded
32
Sources
1
IQ weight
1.15

Score distribution

-1.3
Mean
-0.873
Median
38.1
Std dev
-47.7
10th percentile
46.3
90th percentile
-77.9
Min
-77.958
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.15 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.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data58
2Claude 3.5 Sonnetallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data51.2
3Gemini 1.5 Proallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data46.4
4GPT-4 Turboallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data46.3
5Claude 3 Opusallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data37.8
6Llama 3 70B Instructallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data37
7GPT-4-0125-previewallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data35.8
8yi-largeallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data34.5
9Gemini 1.5 Flashallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data32.1
10Nemotron-4 340B Instructallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data30.5
11Reka Coreallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data27.9
12Yi 1.5 34B Chatallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data25.8
13Claude 3 Sonnetallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data25.5
14DeepSeek Coder V2 Instructallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data20.4
15Qwen2 72B Instructallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data18.5
16GLM-4-9b-chatallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-0.873
17Command R Plusallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-10.1
18Mistral Largeallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307, 2402· WildBench direct data-10.9
19Reka Flashallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-16.1
20Starling LM 7B Betaallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-20.6
21Nous Hermes 2 Mixtral 8x7B DPOallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-21.3
22Mixtral 8x7B Instruct v0.1allenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-22.1
23Llama 3 8B Instructallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-23.6
24Command R+allenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-33.6
25Tulu 2 DPO 70Ballenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-35.4
26GPT-3.5 Turboallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-36.4
27Reka Edgeallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-40.2
28Mistral 7B Instruct v0.2allenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-43.6
29Llama-2-70b-chat-hfallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-47.7
30Llama-2-7b-chat-hfallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-62.6
31Gemma 7B ITallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-65.6
32Gemma 2B ITallenai/WildBench, haiku_reward.Math & Data Analysis.K=-1, claude-3-haiku-20240307· WildBench direct data-77.9

Data credits: WildBench, Allen Institute for AI