Grok 4.3 is a multimodal large language model developed by xAI, officially released on April 30, 2026. It is a reasoning-centric model that employs an always-on chain-of-thought mechanism designed to handle complex logic, mathematics, and scientific analysis. Unlike traditional conversational models, its reasoning capabilities are integrated into every request, prioritizing factual accuracy and multi-step investigation for autonomous agent workflows.
The model features a context window of 1 million tokens, specifically optimized for long-horizon tasks and deep research. In addition to text and image inputs, it introduces native video processing as a first-class input type. Grok 4.3 is also capable of autonomous document generation, allowing it to produce structured files such as PDFs, PowerPoint presentations, and spreadsheets directly from prompts.
Capabilities and Tools
Grok 4.3 includes a built-in server-side code execution environment, enabling it to write, run, and iterate on code to solve problems or generate data files. To bypass its December 2025 knowledge cutoff, the model utilizes real-time search integration across the web and the X social platform. On third-party benchmarks like the Artificial Analysis Intelligence Index, it has demonstrated high proficiency in tool-calling and instruction-following, though it is often positioned as a high-efficiency alternative rather than a total frontier leader.
For developers using the API, a key implementation detail involves the reasoning_details field; preserving this metadata when passing messages back to the model allows it to maintain its internal reasoning state across a conversation. The release was accompanied by a "Custom Voices" suite, which provides low-latency voice cloning and speech synthesis capabilities compatible with the model's output.