Baidu Wenxin large model version comparison test: ERNIE 3.5 8K VS ERNIE 4.0 8K
Origin
The AI SEO WordPress Plugin I developed has recently been downloaded and paid by many users. In the past two days, a friend wants to buy the AI SEO plug-in, but he needs to use Baidu Wenxin large model version 4.0 (ERNIE 4.0 8K), but my AI SEO CN plug-in has been using version 3.5, which is the version ERNIE 3.5 8K .
Therefore, in the past two days, I have been upgrading the CN version of AI SEO WordPress Plugin to support Baidu Wenxin large model version 4.0.
In this process, actual testing of these two versions is indispensable. These tests are summarized here.
ERNIE 3.5 8K VS ERNIE 4.0 8K: Comparison of article generation effects
In the following two articles related to red wine, the prompt words are the same, and the results generated are not much different. At least for SEO there is basically no difference.
I have done a lot of similar articles to generate comparisons, and they basically come to this conclusion. The bigger gap should be in literary talent. The literary talent of articles generated by 4.0 is obviously better than that of 3.5.
ERNIE 3.5 8K VS ERNIE 4.0 8K: Price Comparison
The price of ERNIE 4.0 8K is more expensive than that of ERNIE 3.5 8K.
Entering is more than ten times more expensive. The output is more than twenty times more expensive.
I roughly calculated that 1,000 tokens is about 400 to 500 Chinese characters. If an article is 1,000 Chinese characters (2,000 tokens), then it costs 0.6 yuan to generate an article using ERNIE 4.0 8K. It costs 6,000 yuan to generate 10,000 articles in batches. This does not even include the input price.
If you switch to ERNIE 3.5 8K, it only costs 240 yuan to generate 10,000 articles (only counting output).
From a cost-effective perspective, it is more appropriate to choose ERNIE 3.5 8K.
ERNIE-Bot-turbo is cheaper. In fact, if the outline of the article is well controlled, ERNIE-Bot-turbo is enough for SEO.
ERNIE 3.5 8K VS ERNIE 4.0 8K: Speed
ERNIE 4.0 8K is significantly slower than ERNIE 3.5 8K.
Of course, ERNIE-Bot-turbo is faster
ERNIE 3.5 8K VS ERNIE 4.0 8K: Error Rate
Everything works fine with ERNIE-Bot-turbo and ERNIE 3.5 8K.
ERNIE 4.0 8K has some problems, that is, when generating articles in batches, many returns will be empty. At first, I thought it was because the AI SEO plug-in used concurrent calls to the large model interface, but after changing it to serial, the problem still. Moreover, the phenomenon of returning empty is basically irregular.
Summarize
| Item | ERNIE 4.0 8K | ERNIE 3.5 8K | ERNIE-Bot-turbo |
| Generate Effects | good | Not much different from ERNIE 4.0 8K | Poor, but the article outline is well controlled and is good enough for SEO |
| price | Output: 0.3 yuan/thousand tokens | Output: 0.012/thousand tokens | Output: 0.008/1000 tokens, the cheapest |
| speed | slow | Significantly faster than ERNIE 4.0 8K | Fastest of the three |
| Error rate | A large number of null values are returned during batch generation | Basically no mistakes | Basically no errors occurred |
Overall, for SEO, ERNIE-Bot-turbo and ERNIE 3.5 8K are more suitable.
