Claude 3 overtakes GPT-4 Arena to reach the top! Xiaobei Haiku becomes the new favorite of developers: unrivaled cost-effectiveness

GPT-4 has really been overtaken!

In the large model arena, Claude 3 Grand Cup Opus is the new king, and the Elo score reaches the top of the list.

Even Xiaobei Haiku has entered the second echelon, surpassing the GPT-4-0613 model and leaving GPT-3.5-turbo far behind.

Haiku’s input token price is half cheaper than GPT-3.5-turbo. In terms of output, it is also nearly 2 cheaper than GPT-3.5-turbo per 1 million tokens. bucks.

Compared with GPT-4, the price is only 1/20. And Haiku also supports 200k context.

No wonder some developers bluntly said:

GPT-3.5 is vulnerable to Claude Haiku.

Not only is the price cheaper, Haiku has also gone to the next level.

Before Claude 3 joined the competition, the GPT-4 family had dominated the rankings for almost a year.

Add some background knowledge to new friends: The scoring standard for the LMSYS Chatbot Arena Leaderboard is human scoring, the same question Throw it to two models A and B at the same time, and the human referee will vote for whichever answer is better.

小 Cup Haiku has become the new favorite of developers

It is worth mentioning that in the official congratulatory message from the Arena, the Haiku model was also highlighted:

According to our user preferences, Claude 3 Haiku reaches GPT-4 level.

Its speed, functionality and context length are unique on the market today.

Some netizens have made a direct comparison of how fast it is:

Bolt has not finished running 100 meters, Haiku has already I’ve finished reading the 100k token document...

In fact, with its super cost-effectiveness, in the developer community, Claude 3 Cup Haiku has indeed become a new favorite.

Someone has immediately opened up their imagination and created an open source project that can "overclock" the Haiku effect to the level of a large Opus, attracting a lot of attention.

To put it simply, let Opus be a teacher for Haiku:

First use Opus to generate examples of performing tasks, and then use these teaching cases to improve Haiku posture level.

Key point: Haiku's price is only 1/60 of Opus, and its response speed is 10 times that of Opus.

This project called gpt-prompt-engineer currently has a total of 7.3k stars on GitHub.

It’s this brother Matt who also used Opus and Haiku to create an “AI stock analyst”, which directly became popular on GitHub.

不少應用程式產品也在第一時間接入了Haiku。例如能依據Prompt自動產生網站UI的Vercel。

△就說快不快吧

不過，就在一片給Claude 3新王點讚的聲音中，也有網友認為：

GPT-3.5作為一個「老」模型，至今仍在與最新的模型競爭，這件事本身就很瘋狂了。

但最讓開發者們期待的當然還是：

OpenAI，快起來卷（ doge）。

#

The above is the detailed content of Claude 3 overtakes GPT-4 Arena to reach the top! Xiaobei Haiku becomes the new favorite of developers: unrivaled cost-effectiveness. For more information, please follow other related articles on the PHP Chinese website!