xAI has officially introduced Grok 4.6, positioning its new flagship reasoning model directly alongside GPT-5.6 Sol on the Artificial Analysis Intelligence Index while undercutting competitors on output token costs. Meanwhile, a security paper reveals that encrypted reasoning blocks across OpenAI, Anthropic, and Google APIs contained a fundamental portability flaw, allowing weaker models in the same provider family to decode hidden reasoning traces, API keys, and sensitive user data.

We’ll talk about:

• Grok 4.6 reaching an Intelligence Index score of 61 with a 500K context window, leading on agentic task efficiency while charging $2/M input and $6/M output tokens.
• How security researchers extracted 367 PII artifacts and 182 credentials from 315K+ public reasoning blocks by replaying encrypted traces ...