You guys have been making coding products so competitive, forcing me, old Musk, to step in again!
Musk posted on X, revealing that xAI's own Grok foundation model V9-Medium (1.5T) has completed training.
It is expected to be officially released in about 2 to 3 weeks:
Yes, this new model is clearly making a big push towards programming capabilities.
Musk specifically mentioned that a large amount of Cursor data was added to the supplementary training of V9-Medium, with more to be added later.
What does this mean—
(Probably suggesting, 'Look beyond the neighbors when it comes to coding, check out my house too~')
While hyping the new model, their own new programming product isn't idle either.
xAI's newly released AI programming agent tool Grok Build has also entered the early Beta testing phase, and some users can use it now~
Emm...
It's only been just over a month since acquiring Cursor, and Grok has already started getting its 'bonus meal'.
On one hand, feeding the new model with Cursor data; on the other hand, pushing their own AI programming Agent.
Anyway, Old Musk's Code, better late than never!
Grok's New Model Training is Complete!
Ultimately, this coding stuff is just too competitive...
Looking at the timeline, it's been almost a year since the Grok 4 series, based on the v8-small foundation model, was released.
Actually, the programming capabilities improved a bit with the last generation model, but they just couldn't keep up with how fast the neighbors are iterating!
So, this time xAI didn't choose incremental updates, but saved up for a big move~
Grok's new foundation model, V9-Medium (1.5T), is directly targeting programming capabilities and the developer market.
Even Musk himself couldn't wait to say: The coding ability is much stronger. (doge)
In terms of parameters, V9-Medium has directly reached a scale of 1.5T.
You know, currently all Grok's production traffic is carried by the 0.5T v8-small model, which is only 0.5T in size.
So calculating this way, it's a direct jump to three times the parameters???
The significant increase in parameter scale, of course, also means the model is expected to make considerable progress in deep reasoning capabilities and knowledge reserves.
A larger model capacity often allows it to better understand complex contexts, perform longer chain-of-thought reasoning, 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 prowess—
Old Musk specifically mentioned: xAI intentionally added a large amount of 'Cursor data' to the training of V9-Medium (1.5T).
Everyone should know that Cursor's data isn't an ordinary code repository; it's the real workflow of millions of developers.
Cursor data is more likely to tell the model how developers describe requirements, locate problems, have AI read context, modify files, fix errors, and continue asking questions in real projects.
Consider it a big boost of 'real-world engineering feel' for Grok......
As for when the model will be released?
The official word is that the foundation training of V9-Medium has been fully completed.
Fine-tuning work is in full swing, and the reinforcement learning phase will also start in a few days.
According to xAI's plan, this new model can be officially released to the public in 2 to 3 weeks.
Additionally, it's worth noting—
While promoting their new model, Musk also left an Easter egg in the comments:
The 0.5T model will be open-sourced by the end of this year, it's very practical too!
(You can look forward to it a little~)
AI Programming Tool Grok Build Also Starts Testing
Besides the new model, xAI's own new AI programming tool product has also made new progress.
Currently, the AI programming agent tool Grok Build has entered the early Beta testing phase.
Let me briefly introduce, Grok Build is xAI's own programming Agent and CLI tool.
In form, it's actually closer to terminal programming products like Claude Code, Codex CLI.
Developers can invoke it in a local project directory, letting the AI directly read the project, understand the context, plan changes, modify code files, and so on.
Judging from the official demo video, Grok Build's main promoted capabilities have a strong AI Coding flavor:
Plan Mode: After the user inputs a requirement, Grok Build won't act immediately. Instead, it first formulates a detailed execution plan, lets the user confirm, and then proceeds.
Multiple Subagents Working in Parallel: For large, complex tasks, Grok Build will delegate work to Subagents that run in parallel.
Skills, Marketplace, MCPs Integration: We can easily extend Grok Build's capabilities, invoking various ready-made tools and plugins.
Other Mentioned Capabilities: Supports creating images and videos, building automated workflows, etc.
Regarding specific usage methods, xAI's official documentation also provides three entry points—
One is Interactive TUI.
That is, opening a full-screen interactive interface in the terminal where developers can directly converse back and forth with Grok Build, view plans, and execute tasks.
One is Headless Mode.
More suitable for being embedded into scripts, bots, or automated processes, letting it run tasks in the background according to instructions.
Another is through the Agent Client Protocol to integrate with other applications.
Turning Grok Build into a programming Agent that can be called by external tools.
(Many ways, no shortage of choices~)
On one side, xAI is strengthening the new generation foundation model's code capabilities with Cursor data.
On the other side, launching a programming Agent like Grok Build to directly enter the AI Code tool battlefield.
It seems Musk isn't planning to just compete with Grok in the chatbox this time. (doge)
After Acquiring Cursor, Musk Races to Catch the Coding Tailwind
Old Musk's move of feeding the new model with Cursor data and starting Grok Build testing isn't too sudden.
After all, a few months ago, while xAI was conducting internal layoffs and adjustments, they started heavily engaging with and bringing in Cursor's talent and experience—
Andrew Milich and Jason Ginsberg, Cursor's two core engineering leads who helped grow Cursor from 0 to a valuation of tens of billions of dollars.
Joined xAI one after another, reporting directly to Musk, with the core task of rebuilding Grok's coding capabilities from the ground up.
Then came last month.
Musk stopped playing the talent poaching game and directly targeted Cursor the company itself.
Acquiring Cursor for a price of $600 billion. You should know, this number is more than double Cursor's valuation from November last year…
So looking back, in just one month, Old Musk's Cursor 'stock-up' has already started showing effects on Grok:
Just half a month ago, Grok Build was only just appearing, now it's already starting broad testing in developers' hands.
As for V9-Medium, it was personally named by Musk, stating that a large amount of Cursor data was added to its supplementary training, with more to be added later.
Not only that, in between, Musk also revealed that the new Grok model is being trained normally on the Colossus 2 cluster.
So many actions, so much code vibes.
Therefore, when we look back at the phrase 'more to come' in V9-Medium, it now seems quite thought-provoking.
Translated, I guess it means—
Every extra day Cursor serves customers, Grok gets an extra day of bonus meals.
Next, let's see how Old Musk leads Grok formally into the Code fight, looking forward to it~
References:
[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 public account "Quantum Bit", author: Meng Yao
