01What Grok 4.7 Is and How It Works

Grok 4.7 is the latest AI model from SpaceX AI, released right after Grok 4.6. It is designed for coding, building apps, and handling work tasks like creating documents and presentations. In this episode, Julian Goldie tests it and explains how it performs compared to other top models.

The model works by taking text prompts and generating code, games, or other content. You can access it through the CLI, the API, or tools like Grok Build and Agent OS. It is built to be cheaper and faster than many rivals, making it a good option for everyday projects.

02Main Features of Grok 4.7

Grok 4.7 comes with several key features that stand out. First, it is priced lower than both GPT 5.6 Sol and Fable 5.1 for input and output tokens. That means you can run more tasks without spending as much money.

  • Cheaper tokens – Input and output costs are lower than GPT 5.6 Sol and Fable 5.1.
  • Benchmark improvements – It scores higher than GPT 5.6 Sol on software engineering tasks.
  • CLI and API access – You can use it straight from the command line or through code.
  • Grok Build harness – This tool makes it easier to build and preview creations.
  • Document and presentation skills – It is designed for work tasks, not just coding.

These features make it useful for both developers and non-developers who want a capable AI without high costs.

03How to Set Up Grok 4.7

Getting started with Grok 4.7 is simple, according to Julian. You can use it in several ways, depending on your needs. Here are the basic steps to get it running.

  1. Subscribe to X (Twitter) – The CLI is available if you already have an X subscription.
  2. Get the CLI or API – Use the command line interface or API key for access.
  3. Plug it into Agent OS – Inside the Agent OS workspace, you can switch between Grok 4.7, 4.6, and 4.5.
  4. Use Grok Build – This harness lets you create and preview projects directly.
  5. Save your work – Everything you build gets saved in your workspace for later use.

Julian notes that this setup is easy, even for beginners. You do not need a cloud subscription if you already have X, because the CLI covers access.

04How to Use Grok 4.7 for Building

Julian shows several real examples of using Grok 4.7. One is a 3D world game called Twilight Vale. It is not as polished as Astra 6, but the gameplay is smooth, and building is easy.

Another is a flight simulator, which performed decently. He also shows a game called Neon City.

You can use Grok 4.7 directly in the Agent OS workspace. There, you can preview your builds and switch between models. The process is fast, and the output is almost instant.

Julian also uses it for news gathering. Inside Hermes Agent, there is a tool called Hermes Oracle that pulls the most interesting news from X using Grok on a schedule.

  • Build RPG-style games with simple prompts.
  • Create documents and presentations for work.
  • Pull news from X automatically.

These examples show it is not just for coding. It is a versatile tool for many tasks.

05Grok 4.7 Benchmark Results and Comparisons

In the episode, Julian reviews benchmark data for Grok 4.7. On Cursor Bench 4.0, it scores 46.3%, compared to GPT 5.6 Sol at 41.7%. That is a clear win for Grok.

On Deep SWE v1.1, it scores 71%, while the older Grok 4.6 scored 65%. This shows a big jump.

However, on the Artificial Assist Intelligence Index, Grok 4.7 is far behind models like Deep 6 and Fable 5.1. It is also not compared with Astra 6, so Julian does not think it is close to that level. Still, it is cheaper and performs well on specific tasks.

  • Cursor Bench 4.0: 46.3% vs GPT 5.6 Sol's 41.7%.
  • Deep SWE v1.1: 71% vs Grok 4.6's 65%.
  • Outperforms Fable 5.1 on software engineering.

These results show it is not the best everywhere, but it is strong and affordable.

06Tips, Limits, and Mistakes to Avoid

Julian offers several tips for using Grok 4.7. First, it lacks some detail in coding tests. The builds do not have the same feel as those from Fable 5.1.

So, if you need high polish, this might not be the right model.

Another tip is to set expectations. It is not a replacement for frontier models like Astra 6. Do not expect that level of quality.

But for quick and cheap builds, it is a solid choice.

  • Do not expect Astra 6-level quality.
  • Check that the output lacks fine details.
  • Use it for fast, low-cost experiments.
  • Use the CLI if you already pay for X.

Avoid these mistakes: overpaying for cloud when you have X, or ignoring that it is not top-tier. It is a good tool, but it has limits.

07Who Grok 4.7 Is For and Why It Matters

Grok 4.7 is for anyone who wants a capable AI without spending a lot. If you are a developer or a builder, the cheap tokens and fast performance make it a good fit. It is also useful for non-developers who need to create documents or presentations.

Julian explains that it is a good alternative if you do not want a cloud subscription. Because the CLI works with an X subscription, many people already have access. It also works well with tools like Agent OS and Grok Build, which let you connect all your favorite agents.

Why does it matter? It shows that AI models are getting better and cheaper. Grok releases are coming out every few weeks, and quality keeps improving.

So, this is a practical option for testing ideas and building real things.