SpaceXAI released Grok 4.6 on August 12, delivering Artificial Analysis Intelligence Index parity (score 61) with OpenAI's GPT-5.6 Sol while trailing Anthropic's Claude Fable 5 Max by just one point (62). The model achieves this on the same 1.5 trillion parameter base as Grok 4.5 via post-training improvements: extended supplemental training on curated reasoning data, engineering datasets, and reinforcement learning across coding and knowledge work.
API pricing stands at $2 per million input tokens and $6 per million output tokens, creating an 80-88% cost advantage versus Fable 5 Max ($12/$30 pricing structure) at near-parity performance. Investor Gavin Baker called Grok 4.6 "Pareto dominant." The product is available immediately through Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare, with SpaceX's pending $60 billion acquisition of Cursor (Anysphere) potentially offering access to Colossus's 550,000+ GPU cluster.
The pricing pressure lands at a critical moment: Anthropic targets an IPO as soon as September/October, and OpenAI has filed confidentially, both betting on improving inference-unit economics. Microsoft's $135 billion OpenAI stake and $3.2 billion Anthropic position face valuation risk if Grok's aggressive pricing forces frontier labs to cut margins ahead of their public debuts. For practitioners, the cost-to-capability ratio forces routing decisions—Fable 5 leads on CursorBench and FrontierCode coding benchmarks, Grok 4.6 leads on document understanding and knowledge work (GDPVal, Harvey LAB).