Grok 4.1 Fast benchmark results and details about the Agent Tools API

The xAI team shared benchmark results showing Grok 4.1 Fast as the leader among its competitors. In addition, the developers provided more details about the Agent Tools API infrastructure.

Grok 4.1 Fast takes the leading position

Based on updated benchmark information from xAI, Grok 4.1 Fast takes the leading position in the following categories:

• τ²-bench Telecom: in this category, the new model shows the maximum possible result of 100% and significantly outperforms Claude Sonnet 4.5, GPT-5.1, and Grok 4.
• Berkeley Function Calling v4: the new model scores 72% in the accuracy category at low cost and also takes the lead.
• Long context: the model maintains stable quality up to a 2-million context window. For example, in this category, the new model scores 67%, while Grok 4 scores only 22%.

Agent Tools API – tools for autonomous agents

Agent Tools API is a set of server tools that provide access to external operations and real data.

With the API, an agent can:

• Combine multiple tools in a chain automatically.
• Use intelligent search in a loaded document.
• Connect to external MCP servers.
• Search data in X in real time.
• Run Python code in a secure environment.

A distinctive feature of these tools is that they operate entirely on the xAI infrastructure. This eliminates the need to manage environments, keys, or limits. Grok independently determines the necessary steps and invokes the required tool, and parallel calls can also be implemented.

Currently, there are two model options available: reasoning for tasks requiring deep analysis and non-reasoning for instant responses. The Agent Tools API is available free of charge to xAI users, and the model will be free until December 3.