OpenAI's Code Tool Cursor 'Defects' to Claude, Programmers Applaud

OpenAI heavily invested code tool Cursor has defected:

It quietly switched the default model to Claude.

When we tried to register a new account and install it without modifying any settings, the first time we invoked code completion, it was Claude-3.5-Sonnet that received.

The company behind Cursor, Anysphere, received $8 million from OpenAI at the beginning of its establishment, accounting for 72.7% of the current round of financing. Logically, it should be a die-hard OpenAI camp.

But now it has switched sides, indicating that OpenAI has temporarily lost its dominance in the important scenario of code generation.

Of course, it also reflects that OpenAI's investment agreement is relatively loose.

Or perhaps... they are very confident in their next-generation model and don't have to care about the immediate gains and losses?

Cursor Becomes a Programmer's Top Tool

Cursor has been very popular recently, and many programmer friends have said that they have stopped paying for GitHubCopilot and switched to Cursor.

Note that the price of Cursor is $20 per month, which is twice as expensive as GitHubCopilot next door.

Even the well-known independent developer @levelsio shared that in order to overcome his habit of using VSCode, he staged a cyber version of Your Name:

Changed the name and icon of the Cursor application to VSCode, and the next time he writes code and starts VSCode by muscle memory, it will actually open Cursor.

So, what is the charm of Cursor that makes the programmer community crazy about it?

A large part is due to the ability to use Claude-3.5-Sonnet in it.

In addition to the rich supported models, Cursor also has many notable points in product design.

For example, the new function cursor position prediction just launched in May this year is widely praised. After modifying one place of the code, the AI automatically predicts the next place you want to modify, saving manual navigation.

Another example is the multi-line editing function. The AI suggests modifying multiple lines of code at once to save more time and is often used for format adjustment.

More importantly, the two can be combined, and it becomes:

When you modify the format of one line of code, then just press the tab key a few times, and the format of the entire file will be adjusted.

Recently started testing the experimental Composer function, allowing the AI to edit across files and simultaneously modify all related files in a project, changing many people's work processes.

The official has not yet started promoting it, but the global programmer community has already spread the word, Cursor, you're not promoting this, you're being silly.

Just saw an anecdote that an entrepreneur saw an old-school AI programming guy in a coffee shop with VSCode and Claude windows side by side.

It's really hard to resist not telling him to use Cursor instead.

Now the ending has also been updated: Eventually went to tell him, and now the guy is downloading.

OneMoreThing

The code generation field, Claude is establishing dominance.

In the latest Livebench benchmark test, after the launch of Claude-3.5-Sonnet, it topped the list, and the updated GPT-4oAPI version and ChatGPT version last week failed to achieve a comeback.

Don't forget that there is also the stronger Claude-3.5-Opus that hasn't come out yet.

And at the application level, in addition to Cursor, another online integrated development platform, Replit, has just been promoting the access to Claude-3.5-Sonnet as an important feature.

Reference link:

This article comes from the WeChat public account: Quantum Bits (ID: QbitAI), author: Meng Chen, original title: OpenAI-funded code tool rebels: Defaults to Claude, programmers cheer: Even if it's twice as expensive, they still buy it.