Grok 4.6 Released: Enhanced Agentic AI, Improved Benchmarks, Available in Cursor and Grok Build

xAI has unveiled Grok 4.6, building on Grok 4.5 with enhanced capabilities for long-running agentic tasks, improved handling of complex multi-step processes including research, coding, and transforming ideas into polished applications. The model was trained with an extended phase of additional training using carefully selected model-generated data, improved optimizer, and updated methodology, followed by SFT and RL phases. It demonstrates abilities in self-checking and verification, and excels in converting product ideas into working prototypes, especially in visual and interactive projects. Safety mechanisms were upgraded and calibrated, with the model undergoing the most extensive testing in xAI's history. Benchmark results show Grok 4.6 High achieving AA Intelligence Index 61, GDPVal-AA v2 1753, CursorBench v3.2 69.9%, DeepSWE v1.1 65.9%, FrontierCode v1.1 (Extended) 61.3%, APEX-Agents 57.5%, Terminal-Bench v3.0 26%, APEX-SWE 56.4%, AA-Briefcase 1577, and Harvey LAB (Vals) 15.8%. Grok 4.6 is now available in Cursor and Grok Build, as well as via API and on partner platforms including OpenRouter, Vercel, and Cloudflare. Pricing starts at $2 per million input tokens and $6 per million output tokens, with a faster version costing double. During the first week, the included usage in Grok Build and Cursor is doubled.