xAI has added a new model to its lineup with Grok 4 Fast, a lighter version of its flagship Grok 4 system. The release centers on a clear tradeoff that many AI users care about: keeping performance close to a top model while reducing the amount of compute needed to get the answer.
According to xAI, Grok 4 Fast performs on par with Grok 4 in most tasks, while using about 40 percent less compute. The company says that efficiency can make the price per task drop by as much as 98 percent.
What xAI Is Changing With Grok 4 Fast
Grok 4 Fast is positioned as a smaller, faster and cheaper alternative to Grok 4, but not as a simple downgrade. The central claim is that it can deliver comparable results in many situations while spending fewer resources along the way.
That matters because model performance is not only about final accuracy. For developers, businesses and power users, the cost of reaching a useful answer can be just as important as the answer itself. A model that needs fewer steps and less computation can be easier to use at scale, especially when the same task has to be repeated many times.
xAI highlights a specific efficiency gain around so-called "thinking tokens." Grok 4 Fast uses an average of 40 percent fewer of these tokens while reaching similar results. In plain terms, the model is designed to take fewer intermediate reasoning steps before producing an answer.
The company says the difference is most visible on complex problems. Those are the cases where other models may require more internal steps, which adds computation before the final response appears.
Benchmarks Put It Near Larger Models
The source article lists several benchmark results that show where xAI says Grok 4 Fast stands. On GPQA Diamond, the model scores 85.7 percent. On AIME 2025, it reaches 92.0 percent.
Those scores are described as close to models such as Grok 4 and even GPT-5. The important point is not just that Grok 4 Fast is faster or cheaper, but that xAI is presenting it as competitive with larger or more expensive systems on demanding tests.
Benchmarks do not cover every real-world use case, but they offer a way to compare models across shared tasks. Here, the source emphasizes that Grok 4 Fast is not only being judged on speed or cost. It is also being measured against reasoning-heavy evaluations where accuracy matters.
For users deciding between models, that combination is the main story: Grok 4 Fast is being pitched as a system that can stay close to higher-end models while reducing the computational path required to get there.
A Single Model for Different Workloads
Earlier versions used separate models for simple answers and reasoning-heavy tasks. Grok 4 Fast changes that setup by combining both approaches into one architecture.
The model's behavior is controlled through the system prompt. That means the same underlying model can be directed toward different styles of work, rather than relying on a split between one model for quick responses and another for deeper reasoning.
The source describes this as part of a broader trend toward hybrid models. In practice, the appeal is straightforward: users and developers can rely on one model family across a wider range of tasks, while still choosing how it should behave for a particular job.
xAI is also offering Grok 4 Fast in two versions. One is optimized for reasoning-heavy work, and the other is optimized for quick answers. Both support a 2-million-token context window.
Tool Use Is Part of the Pitch
Grok 4 Fast has also been trained to use external tools on its own. The source specifically names web browsing and code execution as examples.
This matters because some AI tasks require more than a static answer from the model. A system that can browse the web or run code can approach certain problems differently, especially when the answer depends on searching, checking or executing something outside the model itself.
The benchmark results in this area are notable. Grok 4 Fast scores 44.9 percent on BrowseComp and 74 percent on X Bench Deepsearch, where it outperforms Grok 4. In LMArena-Search, it also tops OpenAI's o3-websearch, which previously held the lead.
The source also says Grok 4 Fast currently ranks 8th in Text Arena, ahead of other models in a similar size range. That ranking supports xAI's broader framing of the model as smaller and more efficient without being limited to basic tasks.
Where Grok 4 Fast Is Available
Grok 4 Fast is available through grok.com, the iOS and Android apps, and the xAI API. For users who want to test it without going directly through those channels, the source says it is also free to use for now through OpenRouter and Vercel.
Pricing ranges from $0.05 to $1.00 per million tokens, depending on token type. That pricing structure connects back to the core release message: xAI is trying to make a model that can handle serious work while lowering the cost of each task.
The model's 2-million-token context window is also part of that offering. A larger context window can allow a model to work with more information in a single session, which can be useful for long documents, extended conversations or complex prompts. The source does not detail specific use cases, but the capability is a major part of the product description.
Overall, Grok 4 Fast is a cost and efficiency story as much as a model performance story. xAI is presenting it as a lighter system that can stay close to Grok 4 in most tasks, use less compute, reduce thinking tokens and operate across quick-answer and reasoning-heavy workflows.