Azernews.Az

by Alimat Aliyeva

xAI, the AI company founded by Elon Musk, has announced the release of Grok 4 Fast, a new version of its flagship AI model designed to be faster and more cost-effective than Grok 4, Azernews reports.

According to the information, Grok 4 Fast maintains the same level of accuracy while using 40% fewer compute tokens. This key improvement reduces the cost of query execution by approximately 98% compared to the original model, making it much more accessible for a wide range of applications.

Grok 4 Fast is built on an innovative architecture that combines deep reasoning and fast response (non-reasoning) modes. This hybrid approach, similar to strategies employed by OpenAI (GPT-5) and Anthropic (Claude Opus), allows the model to automatically switch between in-depth comprehension and rapid answer generation. Thanks to this, Grok 4 Fast claimed first place in the LMArena ranking for search tasks and ranked eighth in text response quality, achieving results comparable to or even surpassing its predecessor on real-world benchmarks.

One of the model’s standout technical features is its support for a context window of up to 2 million tokens, significantly expanding its ability to process and analyze large volumes of information. The model was trained using reinforcement learning techniques and is optimized for quick integration with external tools such as code execution and web search.

Grok 4 Fast is already available to end users via the OpenRouter and Vercel AI Gateway platforms, as well as through the xAI API for web, iOS, and Android applications. Pricing starts at $0.20 per 1 million tokens.

With Grok 4 Fast, xAI is pushing the boundaries of AI efficiency and scalability, addressing one of the biggest challenges in the industry — balancing speed, cost, and accuracy. The massive context window could revolutionize how AI handles complex tasks such as long-form content generation, real-time data analysis, and multi-document understanding, potentially setting new standards for next-generation AI models.