Grok 4.6 (high)

Grok 4.6 (high) is a frontier large language model released by SpaceXAI on August 12, 2026. Serving as a post-training refinement of the Grok 4.5 architecture, the model is specifically optimized for long-running autonomous agents, complex software engineering, and high-fidelity knowledge work. It is designed to sustain performance across multi-step tasks without drifting, focusing on efficiency and improved reasoning rather than a traditional increase in parameter scale.

The model utilizes a 500,000-token context window and features an enhanced ability to perform self-verification and error correction during complex workflows. SpaceXAI developed the model using a combination of curated, model-generated reasoning data and high-quality engineering datasets. This training regime included specialized supervised fine-tuning (SFT) and reinforcement learning (RL) in agentic environments, allowing the model to navigate unfamiliar technical domains and iterate on software prototypes from high-level descriptions.

Grok 4.6 introduces a tiered reasoning system, where the "high" configuration serves as a primary setting for professional-grade tasks. It is optimized for "long-horizon" projects, such as researching complex topics or performing repository-wide refactors in coding environments. Performance benchmarks indicate that the model matches leading contemporary models like GPT-5.6 Sol on general intelligence indices while demonstrating particular proficiency in professional knowledge evaluations such as AA-Briefcase and specialized legal tests.

Integrated into the Grok Build development environment and the Cursor platform, the model is oriented toward agentic coding and professional research applications. While it maintains a competitive pricing structure for enterprise API users, it remains a proprietary model with no open-weight equivalent available for self-hosting.