xAI launches Grok 4.1 Fast, new tool-calling model with a 2M context window

On November 19, 2025, xAI launched Grok 4.1 Fast, a groundbreaking addition to its AI lineup, alongside the new Agent Tools API. Described as “our best tool-calling model with a 2M context window,” Grok 4.1 Fast is engineered for speed, accuracy, and real-world enterprise applications, marking a significant shift in autonomous AI agents.

At its core, Grok 4.1 Fast is blazing-fast inference and rapid agentic task completion, making it ideal for demanding scenarios like customer support, finance, and deep research. The model supports an enormous 2-million-token context window, one of the largest available. It ensures consistent performance across its full length, avoiding the degradation common in longer interactions.

xAI achieved this through long-horizon reinforcement learning (RL) focused on multi-turn scenarios and simulated environments, exposing the model to diverse tools across dozens of domains. This training ensures Grok 4.1 Fast handles complex, multi-step reasoning without losing accuracy over extended conversations.

The model halves hallucination rates compared to the previous Grok 4 Fast, while maintaining comparable performance on benchmarks like FActScore. It shines in factuality and tool use, positioning it as “the best agent for deep research” with native integration into the X ecosystem for real-time information.

xAI released two variants:

• grok-4-1-fast-reasoning: Optimized for maximal intelligence in intricate tasks.
• grok-4-1-fast-non-reasoning: Delivers instant responses for speed-critical applications.

New Developments: The Agent Tools API

Grok 4.1 Fast brings Agent Tools API, a suite of server-side tools that transform Grok 4.1 Fast into a fully autonomous agent. Developers no longer manage API keys, rate limits, sandboxes, or pipelines – everything runs on xAI’s infrastructure. Grok autonomously decides tool usage, often invoking multiple in parallel across turns.

Key tools:

• Realtime X and Web Search: For current events, trends, and multihop queries.
• Files Search: Retrieves and cites from uploaded documents.
• Code Execution: Secure Python sandbox for data analysis, simulations, and more.
• Collections Search: Advanced document retrieval.
• MCP Tools: Connects to external servers for custom integrations.

Examples showcased include a hotel booking agent (identifying guests, checking availability, upgrading bookings) and a Tesla Robotaxi sentiment analyzer using X posts, web search, and code execution. This API enables “production-grade agents that specialize in tool calling and agentic search,” simplifying development dramatically.

Benchmarks

• τ²-bench Telecom (agentic tool use in customer support): Perfect 100% score.
• Berkeley Function Calling v4: 72% accuracy.
• Agentic Search Benchmarks (with Agent Tools):
• Research-Eval: 63.9 (outperforming GPT-5’s 45.5).
• Reka FRAMES: 87.6.
• Internal X Browse: 56.3 (far ahead of competitors like GPT-5 at 24.2).

These results, combined with low average costs per query (e.g., $0.046–$0.091), establish Grok 4.1 Fast as state-of-the-art in factuality, efficiency, and real-world agent performance.

Availability and Access

Grok 4.1 Fast and the Agent Tools API are immediately available through the xAI API. To celebrate the launch, xAI made both free for the next two weeks (until December 3, 2025) via partnerships like OpenRouter. Developers can get started by creating an API key at console.x.ai.

Post-promotion pricing remains competitive: $0.20–$0.05 per million input tokens (with caching discounts), $0.50 per million output tokens, and $5 per 1,000 successful tool invocations. Full documentation is at docs.x.ai.

Note that a separate consumer-oriented Grok 4.1 model (focused on emotional intelligence and creativity) was released days earlier, on November 17, available directly to all users on grok.com, X, and mobile apps. However, the “Fast” variant targets API and agentic use cases.