Introducing Grok 4.7: Meet the new AI model from xAI and the latest test results

xAI launched Grok 4.7—the latest version of the Grok AI model—on September 21, 2026. The company describes it as xAI’s most capable model to date for coding and knowledge-based tasks, specifically designed to handle complex assignments that require extensive processing time. 

However, initial feedback from independent testers and users has been mixed. Some test results indicate clear improvements in Grok 4.7 compared to Grok 4.6, while other tests show the model responding more slowly, consuming more resources, and failing to consistently outperform competing AI models. 

So, what has changed in Grok 4.7?

What is Grok 4.7?

Grok 4.7 is a new AI model from xAI, the company that developed Grok.

For general users, the easiest way to understand it is that Grok 4.7 is designed to take time to think through and handle complex problems, rather than rushing to provide an immediate answer.

This model is specifically designed for coding, research, document-related tasks, and other multi-step workflows.

xAI states that Grok 4.7 utilizes a larger foundation than Grok 4.6 and has undergone a longer training process involving more complex tasks. Additionally, it is designed to scrutinize its own output more thoroughly and supports longer conversations or tasks.

How does Grok 4.7 differ from Grok 4.6?

The key difference lies not merely in being a newer generation of chatbot.

Grok 4.7 is designed to handle more complex and time-consuming tasks.

For example, instead of having AI write just a small snippet of code, a user might instruct it to work on a large-scale software project—identifying issues, fixing multiple sections, and verifying that the corrections function correctly.

xAI stated that Grok 4.7 was trained on tasks that could take several hours to execute. It was also trained to better understand the environment in which the Grok bot operates, with the aim of enabling the model to more effectively utilize tools and complete multi-step tasks.

Coding is one of the key areas of focus

Coding is one of the key areas where xAI showcases the capabilities of Grok 4.7. 

According to test results released by xAI, Grok 4.7 scored 46.3% on CursorBench 4.0—up from the 40.4% achieved by Grok 4.6—while Terminal-Bench results also showed significant improvement compared to the previous model. 

Simply put, these tests require AI models to perform actual software development tasks, rather than just answering questions about programming. 

However, benchmark results must be viewed in context, as each type of test measures different capabilities; the comparative results released by xAI do not indicate that Grok 4.7 outperforms every competing AI model in every aspect. 

For example, xAI's own charts show that other AI models performed better in certain tests involving coding and computer-based tasks. 

It Can Work on More Than Code 

Grok 4.7 is not only intended for programmers. 

xAI also highlights work involving documents, presentations, research, legal tasks, and other professional activities. 

The company says Grok 4.7 improved on several knowledge-work tests compared with Grok 4.6. It also reports strong results on some legal and electrical-engineering tests. 

This is part of a broader change in AI. Newer models are increasingly being designed not just to answer questions, but to complete longer tasks using multiple steps. 

What Independent Testing Says 

This is where the picture becomes more complicated. 

Artificial Analysis currently gives Grok 4.7 an Intelligence Index score of 46 and describes it as a strong model with relatively low token pricing. However, it also records a relatively slow output speed and unusually high output volume. 

That last point matters because Grok 4.7 may use many more AI-generated tokens while working on a difficult task. 

In other words, the price per token and the cost of completing a whole task are not necessarily the same thing. 

Several independent reviews have pointed to this issue. Some found Grok 4.7 capable and competitive, while others found that it used significantly more tokens or took longer than expected. 

Early User Reactions Are Mixed 

Early user feedback also does not point in only one direction. 

Some Cursor users reported that Grok 4.7 followed instructions better and produced more thorough results. One hands-on test reported better prompt following and more careful work, but also described the model as heavy on token usage. 

Other users were less impressed. 

In Cursor's community forum, some early testers complained that Grok 4.7 was more verbose and less precise than previous versions. Another discussion argued that the improvement in intelligence did not feel large enough to justify the additional work and waiting time for some tasks. 

These are individual experiences, not controlled scientific measurements, so they should not be treated as proof that the model is universally better or worse. 

What About Speed? 

Speed is one of the more interesting parts of the Grok 4.7 discussion. 

xAI says Grok 4.7 is served at the same standard speed and price as Grok 4.6. There is also a faster version available in certain products, including Cursor and Grok Build. 

But independent measurements show that Grok 4.7 can take a long time to complete difficult tasks, particularly when using higher reasoning settings. 

Artificial Analysis currently lists an output speed of about 56.7 tokens per second for the xHigh version and describes the model as notably slow and verbose. 

For a user asking a simple question, this may not matter much. For someone asking the AI to work continuously on a large project, it can matter considerably. 

Should Everyday Users Care ? 

For someone who simply uses AI to write emails, summarize information, brainstorm ideas, or ask everyday questions, Grok 4.7 may not feel dramatically different from other modern AI models. 

Its main improvements are aimed at more complicated work. 

The difference becomes more apparent when AI is tasked with large-scale software projects, multi-stage research, document creation, or the continuous use of tools over an extended period. 

For these tasks, the ability to keep working, check results, and manage a large amount of information can be more important than simply producing a quick answer. 

The Bigger Story Behind Grok 4.7

Grok 4.7 is part of a larger shift in the AI industry.

AI companies are increasingly moving from chatbots that simply answer questions toward systems that can work on a task for a longer period of time.

That means the next generation of AI is being judged on more than answer quality.

People are also looking at:

• How reliably it completes a task
• How long it takes
• How much information it uses
• How often it makes mistakes
• How much a completed task costs
• How well it works with other software

Grok 4.7 shows both sides of this transition. It demonstrates meaningful progress over Grok 4.6 in several published tests, while early independent testing also shows that stronger performance can come with higher token use and longer processing times.

Conclusion

Grok 4.7 is an important new release from xAI, particularly for coding and longer, more complicated AI tasks.

The early evidence does not support a simple “best AI model” conclusion. xAI's own benchmarks show improvements in several areas, while independent tests and user feedback reveal trade-offs involving speed, verbosity, and how much work the model performs before reaching an answer.

For beginners, the easiest way to think about Grok 4.7 is as a model designed to work harder on difficult tasks rather than simply answer faster.

Whether that extra effort is useful depends on what the user is asking it to do. And because the model is still new, more real-world testing will be needed before its longer-term strengths and weaknesses become clear.