SpaceXAI released Grok 4.6 on August 12, pricing it at $2 per million input tokens and $6 per million output tokens—roughly 80-88% cheaper than Anthropic's Fable 5 Max, which costs $10 input and $50 output. On the Artificial Analysis Intelligence Index, Grok 4.6 scores 61, tied with GPT-5.6 Sol Max and just one point behind Fable 5 Max's 62.
Grok 4.6 is not a new base model but a post-training upgrade on the same 1.5 trillion parameter V9 foundation as Grok 4.5. The improvements came through extended training on curated model-generated reasoning data, high-quality engineering content, improved optimizer recipes, and reinforcement learning in agentic environments covering coding, CAD, and kernel optimization. The model ships with a 500,000-token context window.
The release signals a shift in frontier AI pricing power: three models (Grok 4.6, Qwen 3.8-Max, and DeepSeek V4-Pro) launched within 24 hours, all competing on cost because capability gaps are assumed. This cost convergence at the frontier matters for architects building production agentic systems: model routing strategies and total-cost-per-task economics now outweigh single-model selection.