Grok 4 is here. Everything you need to know about xAI’s most ambitious AI yet

Grok 4 just launched and Elon Musk says it’s smarter than most grad students.

Unveiled during a livestream on July 9, 2025, xAI’s latest chatbot comes with bold promises and bigger ambitions. Musk says Grok 4 can handle everything from coding to complex problem-solving across different fields, and do it faster and better than earlier versions. 

The launch also comes at a tense moment for Musk’s companies, with leadership changes at X and xAI and rising pressure over content moderation. Still, the spotlight stayed on Grok 4 and whether it can hold its own against rivals like GPT-4o, Claude 3, and Gemini.

If you’re wondering what Grok 4 offers, how it differs from previous versions, and how you can start leveraging its features—this post covers everything you need to know.

How much better is Grok 4 than Grok 3?

• Grok 4 provides “scientist-level reasoning.”
• Context length increased from 32,000 tokens (Grok 3) to 130,000 (Grok 4).
• Faster and more accurate, with fewer hallucinations and better GPU optimisation.
• Musk says Grok 4 rivals GPT-5 and Claude 4 Opus and claims they’ve “run out of test questions.”

xAI has been pouring around $1 billion a month into catching up with OpenAI, Google, and Meta. In March, it officially merged with X to pool resources and speed up Grok’s development—and that investment seems to be paying off.

× Sponsored

Grok 4 was trained on xAI’s Colossus supercomputer, which Musk says gives it “scientist-level reasoning.” Compared to Grok 3, the new model can now handle longer prompts—up to 130,000 tokens (up from 32,000). It’s also faster, thanks to better GPU setups, and xAI says it’s more accurate with fewer made-up answers.

During the launch, Musk claimed the xAI team had “run out of test questions” and called reality itself “the ultimate reasoning test.” The message was clear: xAI sees Grok 4 as ready to go head-to-head with models like GPT-5 and Claude 4 Opus.

Is there a Grok 4 for developers?

• Grok 4 comes in two versions: a general model and a developer-focused one.
• The general Grok 4 handles writing, research, translation, math, and reasoning tasks.
• Grok 4 Code is built for developers and integrates with tools like Cursor.

xAI released two versions of Grok 4, and one is built specifically for developers. The general version handles tasks like writing, research, translation, and complex reasoning across math, language, and code.

The developer-focused version, Grok 4 Code, is designed to work inside tools like the Cursor code editor. It helps with code generation, auto-complete, debugging, and even spotting bugs as you go. It’s meant to be a coding assistant for both professionals and hobbyists, competing with tools like GitHub Copilot and GPT-4’s code interpreter.

Early feedback says Grok 4 Code gives smarter suggestions than typical AI coding tools. It seems to better understand your full project, offers help with structure and architecture, and supports deeper integration with your (Integrated Development Environment) IDE, including live file editing and optimisation tips.

What other new features does Grok 4 have?

• Now handles images and promises video soon (multimodal support)
• Voice chats feel smoother, though still not as fluid as Gemini Live
• Keeps live web access with DeepSearch, especially strong for real-time info from X
• Improved at memes, jokes, and internet slang—more “online” than most AI models
• Still trails in image generation compared to Google’s Imagen 3

Grok 4 makes some big upgrades over its previous version, especially in how it handles different types of content. Unlike Grok 3, which was limited to text, Grok 4 can now understand images—and video support is on the way. This puts it in the same league as top AI models like GPT-4o and Gemini 2.5 Pro, though it’s still behind when it comes to image generation, especially compared to Google’s Imagen 3.

Voice conversations are also getting better. Responses feel more natural and less choppy, though they’re still not as smooth as what you’d get with something like Gemini Live.

Grok 4 keeps one of its standout features: DeepSearch, which gives it live access to the web, especially content from X. That means it can pull in real-time news, market data, or trending posts without you having to leave the app or open a browser. Most competitors still limit or block that kind of access.

One thing that really sets Grok apart is how well it understands internet culture. Grok 4 is better than ever at picking up on memes, jokes, and slang. That makes it feel more fluent in how people actually talk online and gives it some personality that other models don’t quite match yet.

Grok 4 vs. Grok 3: Side-by-Side Comparison

How to access Grok 4

If you want to try Grok 4 in chat:

1. Go to the Grok site and log in with your Super Grok account.
2. In the chat window, click the model dropdown (top corner of the chatbox).
3. Select Grok 4 to start using it.

To use Grok 4 via API:

1. Head to x.ai/api and log into the API Console.
2. Click API Keys, then Create API Key.
3. Give your key a name, hit Save, and your Grok API key will be ready.
4. To make API calls with Grok 4, use the endpoint at docs.x.ai and follow the code examples there.

Is Grok 4 worth it?

Based on what we saw in the demo and public details, Grok 4 shows major improvements in reasoning, real-world simulations, and handling live data.

During the launch event, Grok broke down a 100-page real estate document with impressive precision, clearly explaining things like parking allocation and apartment distribution—outperforming many other chatbots in depth and clarity.

In financial tasks, it stood out by pulling live stock data and offering well-explained summaries of market movements—something most competitors avoid or can’t do in real time. It also handled typo detection well, catching 66 out of 72 errors, matching Gemini’s accuracy while offering more detailed explanations.

Beyond that, Grok 4 is being trained to generate video content, and xAI says it will power the first AI-made TV show this year, with a fully AI-generated movie expected next year. Musk even claimed Grok may eventually discover new tech—or new laws of physics.

In business simulations, Grok 4 outperformed all other models by doubling the net worth in a vending machine scenario. In science, it’s helping speed up gene editing research by scanning massive lab datasets. It also excels in fields like radiology (reading chest X-rays), historical analysis, and game development (it helped build a working 3D shooter in 4 hours).

xAI also introduced realistic voice outputs (like a British-accented assistant called Eve) and showcased Grok’s ability to help with programming by debugging entire source code files. Still, it’s not all seamless. Some advanced tasks reportedly take longer to process during peak times, even though xAI advertises a 24-hour turnaround.

How much does Grok 4 cost?

• SuperGrok Heavy costs $300/month or $3,000/year.
• Includes early access to Grok 4 Heavy and upcoming features (coding, multimodal, video).
• API access is live.

xAI launched Grok 4 and Grok 4 Heavy alongside a new ultra-premium plan called SuperGrok Heavy, priced at $300/month or $3,000/year, making it the most expensive AI subscription among major providers. 

The plan gives users early access to Grok 4 Heavy and upcoming tools, including a coding model in August, a multimodal agent in September, and a video-generation model in October. Grok 4 supports a 128K token context, like GPT-4o Plus and Manus AI, while Claude and Gemini go higher. Grok 4 Heavy’s context size is still unknown.

xAI’s enterprise push is only two months old, and recent trust issues haven’t helped. Still, it’s betting on API access and future cloud partnerships to win over developers.

For developers, Grok 4 Code’s specialised focus might justify the premium, especially if IDE integration delivers on promises.

What can’t Grok 4 do yet?

• No memory
• No chat organisation: no folders, tags, or project tools
• Limited to xAI, no plugins or external service links

Grok 4 has improved in speed and intelligence, but it’s still missing key features that hold it back from being a top everyday AI tool.

First, it doesn’t remember past conversations. That means if you leave a session and come back, you have to re-explain everything—unlike ChatGPT, which can follow long-term threads. This makes Grok less useful for research, coding projects, or anything that needs continuity.

There’s also no way to organize your chats. While OpenAI lets users save, label, and return to specific projects, Grok offers no such tools. It’s a one-window-at-a-time experience.

Integration is another weak spot. Grok works mostly as a standalone chatbot. It doesn’t have the plugin support that ChatGPT does or the built-in links to Gmail, Docs, and other tools like Gemini. For anyone managing complex workflows across platforms, that’s a big gap.

Why is Grok 4 so controversial?

Grok’s promise of “free speech AI” has attracted users who feel mainstream chatbots are too restricted, but it’s also led to some serious issues. Ahead of Grok 4’s launch, xAI had to rush out emergency content moderation tools after the bot generated harmful or offensive replies.

One incident involved a Turkish government minister threatening to ban X after Grok used inappropriate language. Around the same time, Grok was also caught repeating antisemitic tropes and dangerous rhetoric, which xAI later blamed on the model being “too compliant” with user prompts. Musk had recently asked the public to help make Grok more “politically incorrect,” but later said the issue was being addressed. 

Musk touched on the topic during the livestream, saying they want Grok to be “a good AI,” but didn’t address any specific problems. The situation highlights the ongoing tension between AI openness and safety—how much freedom is too much?

Should you use Grok on X or the website?

Grok is available in two places: inside X (formerly Twitter) and on a separate site. Both offer similar performance, but using Grok through X gives you more ways to personalise its behavior,kind of like ChatGPT’s “custom instructions” feature.

Using Grok within X also gives it better access to trending posts and live updates, which can be useful if you want real-time info. But that close connection to X is part of why Grok has landed in hot water—it’s harder to control what it says when it’s so closely tied to social media.

What does Grok 4 mean for AI competition?

With GPT-5 on the way and Google doubling down on Gemini, Grok 4 joins a fast-moving race between tech giants. But while others lean into safety and reliability, xAI is taking focusing on speed, personality, and giving developers more room to build.

If Grok 4 delivers on its promises, it could win over advanced users who want smarter coding help, real-time answers, and fewer limits. Still, its unfiltered style means it’ll have to steer clear of the same issues that held earlier versions back.

At its core, Grok 4 feels like a bold move from Musk—unpredictable, packed with potential, and clearly aiming to shake things up.