SpaceXAI Releases Grok 4.6 with 500K-Token Context for Agents and Coding

SpaceXAI released Grok 4.6, a post-training upgrade to Grok 4.5 featuring a 500,000-token context window and enhanced performance on the Artificial Analysis Intelligence Index (scoring 61, up from 56 and tied with GPT-5.6 Sol Max). The model was improved through extended supplemental training on curated reasoning data and engineering examples, regenerated supervised fine-tuning, and reinforcement learning in agentic environments. Available through the xAI API and integrated into Cursor and Grok Build, Grok 4.6 shows improved self-testing and verification on longer reasoning trajectories, though this observation comes from internal testing rather than independent measurement.