xAI releases Grok 4.6, matching GPT-5.6 Sol for world's third-best on Artificial Analysis

Grok 4.6 posts frontier-level results on coding, terminal, knowledge-work, and agent benchmarks, with Musk claiming the upcoming Grok 4.7 will surpass all current models. Reports peg the model at roughly 1.5 trillion parameters, with a fast variant priced at double the standard tokens. It is available in Cursor and via API, positioning xAI directly against OpenAI, Anthropic, and Google for agentic coding workloads.

Grok Bot, opened to public beta on August 11, is a multi-agent system that signs into apps, retains context across tasks, and shares information between agents—an attempt to leapfrog single-agent assistants. Cognition Labs reported that Grok 4.6 inside Devin is a significant jump over 4.5, surpassing GPT-5.6 Sol while trailing only Anthropic's Opus 5 and Fable 5.

Competitively, the third-place Artificial Analysis ranking puts xAI within striking distance of OpenAI and ahead of most Chinese open-weight models on the index, though pricing sits above DeepSeek's aggressive $0.87/M output. Elon Musk posted that 'Grok is the most efficient high intelligence AI.'

Developers welcomed the expanded context window but bristled at the doubled-price fast tier, calling it a 'toll booth,' and noted the coding benchmarks merely match rather than beat GPT-5.6 Sol. The rapid 4.5-to-4.6 cadence also feeds 'model-version-fatigue' chatter. Watch whether Grok Bot's cross-app sign-in raises the same autonomous-agent safety concerns now dominating the week's news.