跪拜 Guibai
← Back to the summary

Grok 4.5 Lands in Cursor, Undercutting Claude on Price for Coding

Hello everyone, I'm Kakarot.

This past week has really been a battle of the gods, with overseas advanced models being released one after another.

The ChatGPT 5.6 model — officials have clearly stated that domestic users will be able to use it as early as tonight, at the latest tomorrow.

This morning, SpaceXAI officially announced the release of the Grok 4.5 model.

Pay attention to the official wording: The first model specifically trained for programming and agents. This indicates that the main improvements this time are in programming and Agent capabilities.

Musk has also been promoting it on his own platform, frantically building hype for Grok.

Third-party professional evaluation agencies have also ranked Grok 4.5. The image below shows the ranking for AI comprehensive intelligence capability.

You can see Grok 4.5 ranks 4th. It is only behind Claude Fable5, Claude Opus4.8, and GPT-5.5. Its capability is still very strong, firmly in the first tier.

The image below is the ranking for programming capability, which is even stronger than Opus 4.8 🤔 You can see Gemini is almost falling off the back of the list.

Google's Gemini 3.5 Pro is about to be released in the next two weeks, so let's look forward to that.

Price?

Someone made a comparison chart: Grok 4.5's score is close to Fable, but the price is only about one-seventeenth of Fable's.

Old Ma is quite honest on this point, personally admitting that the Grok 4.5 model is not as good as Fable5. But he's not wrong either; the vast majority of tasks people use really don't require Fable-level capability.


But what is Grok's advantage? Comparable capability, but cheaper, only 1/7 the price of Fable5.

Grok's model this time was trained jointly by Cursor and SpaceX.

If counting from AI companies, Cursor is already considered an established company. If you've only started learning about AI in recent months, many people probably don't know what Cursor is🤔.

If you've only recently started using AI tools, you might not be as familiar with it, but people in the programmer circle know its weight. But now, Cursor has been acquired by Space X for a price of 60 billion dollars.

Previously, the three major overseas AI giants were also called the "Big Three": OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini. This time, Space X's Grok is also joining the table, so from now on we'll call them the "Big Four"!

So where can you use Grok? 🤔

There are mainly 2 places.

1. Official Grok Build

This is a CLI terminal Agent tool, just like ClaudeCode, without a visual interface.

But it costs 30 US dollars a month, which is roughly 210 RMB. Hmm, still a bit expensive.

2. Cursor (Recommended)

You can use the Grok 4.5 model in Cursor. But you need to subscribe to a Pro membership.

Cursor's membership mechanism is as above. Just get the lowest tier at 20 dollars a month.

It's worth mentioning that Cursor supports Alipay payments, which is very convenient.

If you subscribe to a membership, you'll find that Cursor also has many other models built-in, including GPT 5.5, Claude Fable5, Claude Opus4.8, and 4.6.

Like other Agent products, each model has a quota, meaning you can't use it indefinitely.

But Cursor doesn't seem to have explicitly stated how much you can use. So I asked ChatGPT.

In other words, how much you can use is calculated based on your context content + the model's price 🤔.

What about the actual effect? Just looking at benchmark scores is boring. Let's see what Grok 4.5 can actually make.

First, look at an example from an overseas guy: He used Grok 4.5 to make a rocket tracking application that can display real-time launch data and view a 3D Earth model. This effect is quite something.

The official source also released a universe and solar system simulation scene that can be interacted with in real-time in the browser, dragging, zooming, and accelerating the passage of time, done quite exquisitely.

You can experience it here: https://x.ai/news/grok-4-5

There's another example I find stunning: I had Grok 4.5 generate a 3D fluttering effect of a Five-Star Red Flag based on a real physics engine in Cursor.

When the wind blows, the flag naturally ripples and flutters, not that kind of stiff animation.

Actual testing shows the fidelity of this physics engine is really quite good; the fabric texture and fluttering amplitude are both very realistic.

Time is tight today, so I'll just quickly walk everyone through it. Later, I'll spend time deeply testing the boundaries of Grok 4.5's capabilities.


I'm Kakarot, continuously sharing AI tutorials that are useful to you~ If it's useful to you, give a like and follow before you go 🥰

Comments

Top 1 of 2 from juejin.cn, machine-translated. The original thread is authoritative.

快乐是一切

Waiting for a Jarvis review [onlooker eating melon]

卡卡罗特AI

Which AI is Jarvis? [dazed]