'Grok 4.1 Fast' is here, supporting 2 million token inputs and reducing hallucinations by half compared to the previous generation model.



xAI, the AI company led by Elon Musk, released

Grok 4.1 on Tuesday, November 18, 2025. Additionally, on Thursday, November 20, they announced a new AI model, ' Grok 4.1 Fast, ' and an ' Agent Tools API ' for developers to build AI agents. Grok 4.1 Fast specializes in the AI's ability to function as an 'agent,' autonomously making plans and completing tasks, and has a context window capable of processing an extremely long context of 2 million tokens at a time.

Grok 4.1 Fast and Agent Tools API | xAI
https://x.ai/news/grok-4-1-fast

Models/Grok 4.1 Fast | xAI
https://docs.x.ai/docs/models/grok-4-1-fast-reasoning



Grok 4.1 Fast was designed with a focus on balancing performance, processing speed, and cost efficiency, particularly for 'tool calls,' where AI controls external functions and apps. It is a model that uses reinforcement learning in fields such as customer support and finance, with actual enterprise use in mind. Grok 4.1 Fast underwent reinforcement learning in a simulated environment during development, learning to operate a wide variety of tools across dozens of fields.

In the τ2 -bench Telecom benchmark, which evaluates how well an AI can use tools in scenarios that simulate actual customer support work, Grok 4.1 Fast demonstrated extremely high performance, demonstrating its ability to autonomously select tools and resolve even complex customer support issues.



In addition, the benchmark test called ' Berkeley Function Calling v4 ,' which measures AI's ability to use tools, recorded a 72% accuracy rate, demonstrating high accuracy.



In an internal test called 'X Browse,' which measures the ability to search for information on X (formerly Twitter) and derive answers, Grok 4.1 Fast recorded a score of 56.3, while GPT-5 scored 24.2 and Claude Sonnet 4.5 scored 14.6. Furthermore, in ' Research-Eval Reka ,' which measures in-depth research capabilities, Grok achieved a higher score than GPT-5 while keeping the cost at roughly half.



What's even more noteworthy is that Grok 4.1 Fast is a faster model than Grok 4.1, but it maintains the same performance in FActScore , an index that evaluates factuality, compared to the larger model Grok 4. This shows that even in models that prioritize speed and cost, the accuracy of the information is not sacrificed, xAI said.

Grok 4.1 Fast is designed as a 'Deep Research' agent, and its direct integration into the X ecosystem and powerful web browsing capabilities allow it to search and retrieve information in real time. This ability to access external, up-to-date information and generate answers while corroborating it contributes to a significant reduction in illusions.

The Agent Tools API, which was also announced, allows developers to write just a few lines of code to give Grok functionality such as web searches, X post searches, programmatic code execution, and file upload analysis. Because all of this runs on xAI's infrastructure, developers can avoid the complexities of managing API keys and setting up execution environments.



Usage fees are set at $0.20 (approximately ¥30.00) per million input tokens and $0.50 (approximately ¥75.00) per million output tokens. Cached input, which reduces costs by reusing data once it has been read, costs $0.05 (approximately ¥7.50) per million tokens. Pricing for using the tool starts at $5 (approximately ¥750) per 1,000 successful calls. To celebrate the release, a free trial is available through OpenRouter until December 3, 2025.

Related Posts:

in AI,   Web Service, Posted by log1i_yk