Grok 4.1 Fast and Agent Tools API

xAI unveiled Grok 4.1 Fast and the Agent Tools API, presenting a next-generation agentic setup specialized for tool calling and agentic search. Grok 4.1 Fast is a model that supports a 2M context window, aiming for fast and accurate reasoning and execution in real-world tasks like customer support and finance.

The core focus is real-world tool-use performance. The model underwent RL training across a variety of tools in simulated environments, and was introduced as showing strong performance on τ²-bench Telecom. It also emphasizes performance on the Berkeley Function Calling v4 Benchmark and multi-turn performance in long contexts, explaining that it addresses the weakness of existing agents whose performance degrades as context grows longer.

The Agent Tools API announced alongside it is a bundle of server-side tools that agents can use directly on top of xAI's infrastructure. The features provided are as follows.

• Search Tools: Real-time X and internet search
• Files Search: Search and citation of uploaded documents
• Code Execution: Python execution in a secure sandbox
• MCP Tools: Integration with external MCP servers

The example code shows a setup where web_search(), x_search(), code_execution(), collections_search(), and mcp() are attached to the grok-4-1-fast-reasoning model, calling multiple tools in parallel within a single flow. xAI emphasized that developers no longer need to handle API key management, rate limit handling, sandbox building, or retrieval pipeline operation themselves.

There are two released versions. grok-4-1-fast-reasoning is for maximum intelligence, and grok-4-1-fast-non-reasoning is for instant responses. Pricing starts at $0.20 / 1M tokens for input, $0.05 / 1M tokens for cached input, $0.5 / 1M tokens for output, and $5 / 1000 invocations for tool calls based on successful calls.