xAI introduces Grok 4 Fast, a more efficient AI

xAI has introduced Grok 4 Fast, a model designed to deliver performance close to Grok 4 with less reasoning time and a much lower cost. It is already available to Grok users and developers through the xAI API.

The idea is simple: AI should respond quickly when a question is easy and spend more time thinking when a problem requires it, without having to switch models. Grok 4 Fast combines both modes in a single architecture.

More capability with fewer tokens

Tokens are the units of text that AI processes. The more it uses, the more a query can cost and the longer it can take to respond. According to xAI, Grok 4 Fast uses 40% fewer reasoning tokens on average than Grok 4 to achieve comparable performance in its evaluations.

The company says that this efficiency, combined with a much lower price per token, makes it possible to reduce the cost of achieving the same performance as Grok 4 by 98% in benchmark tests. This figure refers to the cost of obtaining equivalent results, not to an identical reduction across all queries.

In tests published by xAI, Grok 4 Fast outperforms Grok 3 Mini in several math, science, programming and reasoning tasks. It also comes close to Grok 4 in several results:

• 92% on AIME 2025, compared with 91.7% for Grok 4.
• 93.3% on HMMT 2025, above Grok 4's 90%.
• 80% on LiveCodeBench, compared with 79% for Grok 4.
• 85.7% on GPQA Diamond, compared with 87.5% for Grok 4.

These are xAI's internal results and serve as a reference, not a guarantee that the model will get that percentage of answers right in a normal conversation.

Web and X searches

Grok 4 Fast has also been trained to decide when it needs to use external tools, such as web browsing or code execution. It can move between links, consult X posts, analyze images and videos hosted there and combine the information into an answer.

In the search tests disclosed by xAI, it scored 44.9% on BrowseComp, 95% on SimpleQA and 58% on X Browse. The company also says that its search version took first place in LMArena's Search Arena, with 1163 Elo points, 17 more than o3-search.

These figures are especially relevant for queries about current events. If you ask about something that happened recently, the model can search for information instead of relying solely on what it learned during training.

What changes for you

Grok 4 Fast is available to all Grok users, including those on the free plan, according to xAI. In Fast and Auto modes, it should improve searches and queries that require gathering information in particular.

In Auto mode, difficult questions can be handled by Grok 4 Fast without requiring you to manually choose a model. Simple queries will receive quick answers, while more complex ones can trigger longer reasoning.

For developers, xAI offers two variants through its API:

• grok-4-fast-reasoning, designed for tasks that need more analysis.
• grok-4-fast-non-reasoning, intended for direct answers and lower latency.

Both have a context window of 2 million tokens, meaning they can work with a very large amount of text in a single request. This can be useful for analyzing long documents, complete histories or large amounts of information without splitting them into as many fragments.

API pricing starts at $0.20 per million input tokens and $0.50 per million output tokens when a request uses fewer than 128,000 tokens. Above that threshold, the price rises to $0.40 per million input tokens and $1 per million output tokens. Cached input tokens cost $0.05 per million.

Grok 4 Fast shows where the competition is heading: not only toward more capable models, but toward models that can deliver that capability quickly and cheaply. The question now is whether its advantage in testing and pricing holds up in everyday use, especially when searches, tools and real-time information come into play.