Grok 4.3 (medium)

Grok 4.3 (medium) is a reasoning-focused language model developed by xAI, released on April 30, 2026. As a flagship iteration in the Grok series, it is designed for agentic workflows, long-document analysis, and tasks requiring high factual accuracy. The model introduces an improved architecture over its predecessor, Grok 4.20, while maintaining a focus on reducing hallucinations and enhancing instruction-following capabilities. The "medium" designation refers to its parameter size class, positioning it as a balanced option between efficiency and high-level reasoning performance.

The model features a 1-million-token context window, allowing for the processing of extensive datasets, entire codebases, or complex multi-step research tasks. It is natively multimodal, supporting inputs in text, image, and video formats. A core architectural feature is its configurable reasoning mode, which allows users to adjust the model's "thinking" effort across none, low, medium, and high levels. This enables the model to perform deeper internal chain-of-thought processing before providing a final response.

Key Capabilities and Technical Details

Grok 4.3 (medium) includes native support for function calling and structured outputs, facilitating integration into external toolchains and automated systems. It also ships with a built-in code-execution environment that allows the model to write, test, and run code directly during the reasoning process. Its knowledge cutoff is documented as December 2025, providing context for modern events and software libraries up to that period.

Benchmarks indicate that the model is particularly effective for agentic tasks, where it often outperforms larger non-reasoning models in planning and executing multi-step instructions. While the model is optimized for logic and factual retrieval, its "always-on" reasoning capabilities significantly improve its performance in complex coding refactors and mathematical problem-solving compared to standard autoregressive models.