Grok's latest model gets an extra boost from Cursor. Elon Musk: There has been a huge improvement in coding.
I've made your Code products so competitive. You're forcing me, Elon Musk, to step in again!
Elon Musk posted on X, revealing that xAI's in - house Grok base model V9 - Medium (1.5T) has completed its training.
It is expected that in about 2 to 3 weeks, it will be officially released to the public:
Yes, this new model is clearly focused on improving programming capabilities.
Musk specifically mentioned that a large amount of Cursor data was added to the supplementary training of V9 - Medium, and more will be added in the future.
What does this mean?
(It's probably saying, "When it comes to coding, don't just look at the other two companies. Take a look at ours too!")
While the new model is being hyped up, xAI's in - house programming product isn't idle either.
The newly released AI programming agent tool Grok Build by xAI has now entered the early Beta testing phase, and some users can start using it!
Well...
It's only been a little over a month since the acquisition of Cursor, and Grok is already getting an "extra boost".
On one hand, it uses Cursor data to train the new model, and on the other hand, it promotes its own AI programming agent.
Anyway, Elon Musk's Code efforts, though late, have finally arrived!
The new Grok model has completed training!
After all, the competition in coding is extremely fierce...
Looking at the timeline, it's almost been a year since the release of the Grok 4 series based on the v8 - small base model.
Actually, the programming capabilities of the previous - generation model also improved to some extent, but it can't keep up with the rapid iterations of the other companies!
So, this time, xAI didn't take small steps. Instead, it prepared a big move!
Grok's new base model V9 - Medium (1.5T) is directly targeted at programming capabilities and the developer market.
Even Musk himself couldn't wait to say: The programming capabilities are much stronger. (doge)
In terms of parameters, the parameter scale of V9 - Medium has directly reached 1.5T.
You know, currently, all the production traffic of Grok is carried by the 0.5T v8 - small model, which is only 0.5T in size.
So, calculated like this, it's directly a three - fold increase in parameters?
The significant increase in the parameter scale, of course, also means that the model is expected to make considerable progress in deep reasoning ability and knowledge reserve.
A larger model capacity often enables it to better understand complex contexts, conduct longer chain - thinking, and handle more challenging real - world development tasks.
However!
The most interesting part of this new model isn't in the parameters but in its programming skills:
Elon Musk specifically said that xAI added a large amount of "Cursor data" to the training of V9 - Medium (1.5T).
You should know that Cursor's data isn't an ordinary code library. It's the real - world workflows of millions of developers.
Cursor data is more likely to tell the model how developers describe requirements, locate problems, make the AI read the context, modify files, fix errors, and then continue to ask questions in real projects.
It's like giving Grok a big boost in "real - world engineering experience".
As for when the model will be released?
Officially, the basic training of V9 - Medium has been completed.
The fine - tuning work is in full swing, and the reinforcement learning phase will start in a few days.
According to xAI's plan, this new model will be officially released to the public in 2 to 3 weeks.
Moreover, it's worth mentioning that:
While promoting his new model, Musk also left an Easter egg in the comment section:
The 0.5T model will be open - sourced by the end of this year. It's also very useful!
(Everyone can look forward to it a little!)
The AI programming tool Grok Build has also started testing
Besides the new model, xAI's in - house AI programming tool has also made new progress.
Currently, the AI programming agent tool Grok Build has entered the early Beta testing phase.
Let me give you a brief introduction. Grok Build is xAI's in - house programming agent and CLI tool.
In form, it's more similar to terminal programming products like Claude Code and Codex CLI.
Developers can invoke it in the local project directory, allowing the AI to directly read the project, understand the context, plan changes, modify code files, etc.
Judging from the official demonstration video, Grok Build's main capabilities this time are full of AI coding flavor:
Plan mode: After the user inputs requirements, Grok Build won't act immediately. Instead, it will first formulate a detailed execution plan and execute it after the user confirms.
Parallel work of multiple sub - agents: For large and complex tasks, Grok Build will assign the work to parallel - running sub - agents.
Integration of Skills, Marketplace, and MCPs: We can easily expand the capabilities of Grok Build and call various existing tools and plugins.
Other mentioned capabilities: It supports creating images and videos, building automated workflows, etc.
In terms of specific usage methods, xAI's official documentation also provides three entry points:
One is the interactive TUI.
That is, open a full - screen interactive interface in the terminal. Developers can directly communicate with Grok Build, view plans, and execute tasks.
One is the headless mode.
It's more suitable for being integrated into scripts, robots, or automated processes, allowing it to run tasks in the background according to instructions.
The other is to connect to other applications through the Agent Client Protocol.
Turn Grok Build into a programming agent that can be called by external tools.
(There are many ways to choose from!)
On one hand, xAI is using Cursor data to enhance the code capabilities of the new - generation base model.
On the other hand, it's promoting the programming agent Grok Build and directly entering the AI Code tool battlefield.
It's obvious that Musk doesn't plan to just compete in the chat box with Grok this time. (doge)
After acquiring Cursor, Musk is riding the Coding wave
Elon Musk is using Cursor data to train the new model and testing Grok Build.
Actually, it's not that sudden.
After all, a few months ago, while xAI was making internal lay - off adjustments, it started to extensively contact and introduce Cursor's talents and experience:
Andrew Milich and Jason Ginsberg, the two core engineering leaders of Cursor, who helped Cursor grow from scratch to a valuation of billions of dollars, have joined xAI one after another and report directly to Musk. Their core task is to reconstruct Grok's coding capabilities from the ground up.
Then, last month.
Musk stopped poaching talents and directly targeted the Cursor company itself.
He acquired Cursor at a price of 60 billion US dollars. You know, this figure is more than twice the valuation of Cursor in November last year...
Looking back, in just one month, Musk's "acquisition" of Cursor has already shown results on Grok:
Half a month ago, Grok Build was just emerging. Now, it's starting to be widely tested by developers.
As for V9 - Medium, Musk himself mentioned that a large amount of Cursor data was added to the supplementary training, and more will be added in the future.
Moreover, during this period, Musk also revealed that the new Grok model is being normally trained on the Colossus 2 cluster.
With so many actions and so much coding significance.
So, looking back at the phrase "more to come" in V9 - Medium, it now seems quite thought - provoking.
I guess it means:
The more days Cursor serves users, the more "extra boost" Grok gets.
Next, let's see how Musk leads Grok to enter the Code battlefield. I'm looking forward to it!
Reference links:
[1]https://openrouter.ai/x-ai/grok-build-0.1?utm
[2]https://x.com/elonmusk/status/2058787384364265734
[3]https://x.com/JasonBud/status/2058974659648123084
[4]https://x.com/xai/status/2058973760708091907
This article is from the WeChat official account "QbitAI", author: Mengyao. It is published by 36Kr with authorization.