CtrlK
Grok 4.1 Fast is a tool-calling model from xAI announced on November 19, 2025. xAI presented it alongside the Agent Tools API and described it as the company's best tool-calling model, built for enterprise tasks such as customer support, finance, and agentic search.[1]
The launch positioned Grok 4.1 Fast as a reasoning model with a 2 million-token context window and a strong emphasis on multi-turn tool use rather than chat personality.[1][2]
| Area | Details |
|---|---|
| Announcement date | November 19, 2025 |
| Context window | 2 million tokens |
| Stated specialty | Tool calling and agentic search |
| Training emphasis | Reinforcement learning in simulated tool environments |
| Common API model name in docs | grok-4-1-fast-reasoning |
xAI said it trained the model on a wide variety of tools across many domains so that it could stay accurate and efficient even on long-horizon, multi-turn tasks.[1][2]
| Evaluation | Result cited by xAI |
|---|---|
| tau2-bench Telecom | 100% |
| Berkeley Function Calling v4 | 72% overall accuracy |
The company also emphasized that performance held up across long contexts and multi-turn sessions, which it presented as a major issue for other agentic models.[1]
Grok 4.1 Fast was designed to work with xAI's hosted tools for web search, X search, code execution, collections search, and remote MCP connections. xAI's launch article says the model can invoke multiple tools in parallel and continue across several turns until it has enough evidence to produce a final answer.[1][3]
Public xAI documentation and examples use Grok 4.1 Fast through the xAI API and the OpenAI-compatible Responses API surface. The docs also show the model on regional endpoints and in asynchronous, streaming, and editor workflows.[2]