Grok 4 Fast, xAI’s latest model more efficient and economical

xAI has announced Grok 4 Fast, a new version of its language model that seeks to maintain Grok 4's performance with greater efficiency in token usage. According to the company, Grok 4 Fast uses 40% fewer tokens compared to its predecessor, translating into cost reductions of up to 98% in their internal benchmarks.

The model incorporates a 2-million-token context window, allowing it to handle large volumes of information in a single workflow. Its unified architecture combines extended reasoning modes and quick responses in one model, controlled through system instructions, without needing to alternate between different configurations.

Grok 4 Fast includes native tool integration capabilities, trained with reinforcement learning. It can perform web and X (formerly Twitter) searches, as well as process multimedia content, including images and videos, to synthesize information in real time. Additionally, the model allows controlled code execution according to system instructions.

Two variants are offered for developers: grok-4-fast-reasoning and grok-4-fast-non-reasoning, both with the 2-million-token context window. The model is available on the grok.com platform, including all users, and can also be integrated through OpenRouter, Vercel AI Gateway, and xAI API.

In independent LMArena evaluations, Grok 4 Fast reached first position in the Search Arena with 1163 Elo points and eighth place in the Text Arena, showing efficiency and performance comparable to Grok 4 in different test scenarios.

With this update, xAI seeks to offer a language model that combines efficiency, unification of reasoning modes, and search capabilities, maintaining performance standards similar to its previous version but with lower token consumption and greater accessibility for developers and users.

Key points

• Grok 4 Fast reduces operational costs by 98% while maintaining Grok 4's performance
• The model improves token efficiency by 40% compared to its predecessor
• Incorporates unified architecture combining extended reasoning and quick responses in one model
• Includes a 2-million-token context window to handle large information volumes
• Offers native web and X search capabilities with multimedia content processing
• Available to all grok.com users, including free users
• Presented in two variants for developers: grok-4-fast-reasoning and grok-4-fast-non-reasoning
• Achieved first position in LMArena Search Arena with 1163 Elo points