AUGUST 14, 2026 — GROK 4.6 LAUNCH · AA INDEX 61 · $2/M UNDER 200K TOKENS
- Grok 4.6 released August 12: AA Intelligence Index 61 — ties GPT-5.6 Sol Max, one below Fable 5 Max (62). Price: $2/$6/M under 200K tokens. Context: 500K. CursorBench v3.2: 69.9% (vs 4.5's 66.7%). APEX-Agents leader. DeepSWE: GPT-5.6 Sol still leads. Knowledge cutoff: February 1, 2026. Available: xAI API, Grok Build, Cursor, OpenRouter, Vercel, Cloudflare. Pricing trap: doubles to $4/$12/M above 200K tokens — entire request repriced. Grok 4.7 (2.1T) coming weeks. Grok 5 before end 2026. Full review + benchmark table →
Grok 4.6 — The Practical Summary
Per APIdog's full analysis, Grok 4.6 is the same 1.5T parameter base as Grok 4.5 with an upgraded post-training run — not a new architecture. The gains come from better RL in agentic environments and higher-quality engineering data. The result: stronger performance on long multi-step agent tasks, with the model self-verifying its work more on longer trajectories. The AA Index tie with GPT-5.6 Sol (both at 61) is significant — it means Grok 4.6 is now at the frontier tier at $2/M input, compared to GPT-5.6 Sol at $5/M. That's roughly 60% cheaper input for equivalent composite benchmark performance. The caveat: GPT-5.6 Sol still leads on DeepSWE (repository-scale coding), and Grok 4.6's long-context pricing (doubles above 200K tokens) narrows the cost advantage for large-context workloads. Full review →
This Week's Full Coverage
<- August 13: Anthropic Q2 first profit, OpenAI S-1 imminent