xAI Launches Grok Build 0.1 in Public Beta with Enhanced API Features

xAI has announced the public beta availability of Grok Build 0.1 through its official API, marking a significant step in providing developers with advanced AI-driven coding capabilities. Grok Build 0.1 is positioned as xAI's fastest coding model, specifically engineered to excel in agentic coding tasks, web development, and debugging. It also serves as the underlying engine for the Grok Build Command Line Interface (CLI) and is designed for general-purpose tool calling. The model boasts impressive speed, processing over 100 tokens per second, and is priced competitively at $1 per million tokens for input and $2 per million tokens for output. Accompanying the Grok Build 0.1 release are several key API enhancements aimed at improving developer experience and application performance. The new WebSocket Responses API mode allows for driving the Responses API over a single, long-lived WebSocket connection. This innovation is crucial for achieving lower end-to-end latency, particularly for agent workloads that heavily rely on tool interactions. Furthermore, xAI has introduced the Context Compaction API. This feature enables developers to condense lengthy conversations into a shorter context, which can then be reused in subsequent requests. The benefits include reduced costs, faster time-to-first-token, and more precise responses in extended agent loops, addressing a common challenge in managing conversational AI. Another notable addition is the Smart Turn end-of-turn detection for the streaming Speech to Text API. By leveraging a machine learning model, Smart Turn predicts when a speaker has completed their thought at silence boundaries. This significantly reduces false endpointing during dictation, number sequences, and pauses within sentences, leading to smoother and more accurate speech-to-text transcription. These combined updates underscore xAI's commitment to evolving Grok Build into a comprehensive and powerful platform for AI-assisted development.