SpaceXAI's Grok 4.6 dominates agentic AI with $2/$6 pricing

SpaceXAI has unveiled Grok 4.6, marking its first flagship model under the SpaceXAI name and a strategic pivot toward developer and enterprise tools. Built for complex, multi-step agentic tasks—such as researching topics, navigating entire codebases, or prototyping applications—the model introduces enhanced self-testing and verification behaviors. Grok 4.6 demonstrated stronger performance in visual and interactive projects compared to its predecessor, Grok 4.5, and was trained using a longer supplemental run with curated, model-generated data focused on reasoning and technical concepts.

The model is now available in Cursor, the Grok Build tool, and via API, with pricing starting at $2 per million input tokens and $6 per million output tokens. Early benchmarks show Grok 4.6 matching OpenAI's GPT-5.6 Sol on the Artificial Analysis Intelligence Index with a score of 61, and achieving 65.9% on the DeepSWE v1.1 software engineering benchmark—an 11.9-point improvement over Grok 4.5. While competitive, it still trails GPT-5.6 Sol Max (73%) and Anthropic's Fable 5 Max (62) in some tests. xAI is offering double usage in Grok Build and Cursor for the first week to drive adoption.