2024-06-15
数据分析
00

目前看大模型厂家各自最好的模型的API,DeepSeek-V2的性价比遥遥领先

相对性能是用模型在榜单中的得分除以GPT-4-Turbo-0409的得分;有Arena Elo得分的只用Arena Elo,没有Arena Elo有OpenCompass就用OpenCompass,这两都没有的用SuperCLUE。

性价比计算用相对性能分数除以输出价格,因为使用API一般输出token数量要远多余输入。

序号模型名称公司/提供方输入价格(元/百万tokens)输出价格(元/百万tokens)Arena EloOpenCom-pass客观综合OpenCom-pass主观综合SuperCLUE相对GPT-4-Turbo-0409性能性能/输出价格
1DeepSeek-V2深度求索1256.544.50.8938050.446903
2Qwen1.5-72B阿里云510114754.537.568.040.9132170.091322
3Moonshot-v1 -8k月之暗面121253.743.170.420.8566370.071386
4Qwen1.5-110B阿里云714116356.847.40.9259550.06614
5yi-large零一万物2020123974.290.9864650.049323
6abab6.5MiniMax303057.845.80.9168140.03056
7Spark3.5 Max讯飞星火303050.348.169.430.8707960.029027
8Llama-3-70b-Instruct未提供278211530.9179940.011195
9Baichuan4百川10010080.641.0470010.01047
10Mistral-largeMistral29.888.6115353.428.80.9179940.010361
11GPT-4oOpenAI36.2108.6128764.251.11.0246820.009435
12混元-pro腾讯3010072.120.936380.009364
13GLM-4智谱AI100100117557.844.272.580.935510.009355
14Claude 3OpusAnthropic74.47108.6124660.548.174.470.9920380.009135
15Qwen-Max阿里云40120118655.65072.450.9442680.007869
16ERNIE 4.0-8K百度12012054.746.671.90.896460.007471
17Gemini 1.5 Pro谷歌50.715212480.9936310.006537
18GPT-4-Turbo-0409OpenAI72217125663.149.977.0210.004608
19GPT-4-Turbo-1106OpenAI72217125162500.9960190.00459
20Claude 3 SonnetAnthropic21.7108.61199
21Claude 3 HaikuAnthropic29
22abab6.5sMiniMax1010
23abab6.5gMiniMax55
24GPT-4-0613OpenAI217434
25GPT-3.5OpenAI1114
26qwen-Long阿里云0.52
27qwen-turbo阿里云26
28qwen-plus阿里云412
29qwen-max-0428阿里云40120
30qwen-max-0403阿里云40120
31qwen-max-0107阿里云40120
32qwen-max-1201阿里云120120
33qwen-max-longcontext阿里云40120
34Baichuan3-Turbo百川1212
35Baichuan3-Turbo-128k百川2424
36Baichuan2-Turbo百川88
37Baichuan2-Turbo-192k百川1616
38Baichuan2-53B百川2020
39ERNIESpeed百度00
40ERNIELite百度00
41yi-large-turbo零一万物1212
42yi-large-rag零一万物2525
43yi-medium零一万物2.52.5
44yi-medium-200k零一万物1212
45yi-spark零一万物11
46yi-vision零一万物66
47混元-lite腾讯00
48混元-standard腾讯4.55
49混元-standard-256k腾讯1560
50Mixtral 8x22B未提供1443
51Moonshot-v1 -128k月之暗面6060
52Moonshot-v1 -32k月之暗面2424
53GLM-4-0520智谱AI100100
54GLM-4-Air智谱AI11
55GLM-4-Airx智谱AI1010
56GLM-4-Flash智谱AI0.10.1
57GLM-4V智谱AI5050
58GLM-3-Turbo智谱AI11
59Doubao-pro-4k字节跳动0.82
60Doubao-pro-32k字节跳动0.82
61Doubao-pro-128k字节跳动59
62Doubao-lite-4k字节跳动0.30.6
63Doubao-lite-32k字节跳动0.30.6
64Doubao-lite-128k字节跳动0.81
65Doubao-embedding字节跳动0.50.5

本文作者:tsingk

本文链接:

版权声明:本博客所有文章除特别声明外,均采用 BY-NC-SA 许可协议。转载请注明出处!