xAI ships Grok 4.6 with 1.5T params and 500K context, lands in GitHub Copilot

Grok 4.6 is a post-training upgrade over Grok 4.5, a 1.5-trillion-parameter model refined with a longer supplemental training run, regenerated supervised fine-tuning trajectories, and reinforcement learning in agentic environments. It targets long-running agents, coding, and knowledge work with a 500,000-token context window, and scored 61 on the Artificial Analysis Intelligence Index — matching GPT-5.6 Sol and just behind Claude Fable 5.

Distribution was immediate and broad: Grok 4.6 shipped simultaneously across Cursor, Grok Build, GitHub Copilot (Microsoft), and API partners OpenRouter, Vercel, and Cloudflare, priced at $2/$6 per million input/output tokens. Perplexity CEO Aravind Srinivas said his team benchmarked it as an orchestrator on their Wide-And-Deep-Research benchmark and found it 'neatly sits on the Pareto frontier of performance vs cost,' making it available to Pro and Max users.

The competitive story is cost-per-intelligence. Community analysis showed Grok 4.6 and GPT-5.6 Sol tie at 61 overall but split sharply: Sol dominates engineering (DeepSWE 73% vs 65.9%, Terminal-Bench 34.6% vs 26%) while Grok wins professional knowledge work at roughly a fifth of Sol's output cost ($6 vs $30). That positions Grok as the value pick for agentic knowledge tasks where raw engineering benchmarks matter less.

Elon Musk publicly endorsed 4.6 for speed and cost while conceding a competitor is 'smarter overall,' and teased that Grok 4.7 'has a good chance of exceeding all current models in intelligence' — setting expectations for a rapid follow-up. The GitHub Copilot integration is the most consequential distribution win, putting Grok in front of millions of developers inside a Microsoft-owned surface.