Viral Wire

xAI's Grok 4.6 rivals GPT-5.6 at 60% lower price, 500K context

Grok 4.6 ties GPT-5.6 Sol at AA Index 61, priced 60% lower at $2/M.

Deep Dive

xAI officially launched Grok 4.6 on August 12, 2026, and it immediately lands at the frontier of AI benchmarks. The model scores 61 on the AA Intelligence Index, tying OpenAI's GPT-5.6 Sol and sitting just one point behind Fable 5 Max at 62. That parity is remarkable given the pricing: Grok 4.6 charges $2 per million input tokens and $6 per million output tokens for contexts under 200K, versus GPT-5.6 Sol's $5 per million input—a roughly 60% cost advantage for equivalent composite performance. The model also delivers a 500K-token context window and an improved 69.9% on CursorBench v3.2, up from Grok 4.5's 66.7%, while topping the APEX-Agents leaderboard for multi-step reasoning tasks.

Under the hood, Grok 4.6 is not a new architecture—it retains the same 1.5T parameter base as Grok 4.5—but xAI upgraded the post-training pipeline with better reinforcement learning in agentic environments and higher-quality engineering data. This yields stronger self-verification on long trajectories, making it ideal for complex, repository-scale coding and agent workflows. However, there's a pricing trap: any request exceeding 200K tokens gets repriced at $4/$12 per million, narrowing the cost edge for large-context use cases. Grok 4.6 is now live across xAI API, Grok Build, Cursor, OpenRouter, Vercel, and Cloudflare. xAI also confirmed Grok 4.7 (2.1T parameters) is coming in weeks, with Grok 5 due before the end of 2026.

Key Points
  • AA Intelligence Index 61 ties GPT-5.6 Sol, one point below Fable 5 Max at 62
  • Input pricing at $2/M under 200K tokens, doubling to $4/M above that threshold
  • CursorBench v3.2 improved to 69.9% from 66.7%; remains leader on APEX-Agents

Why It Matters

Frontier-level agentic performance now available at 60% lower input cost, forcing OpenAI to justify pricing.

📬 Get the top 10 AI stories daily