xAI releases Grok 4.6 with 500K context, matching GPT-5.6 Sol

SpaceXAI (formerly xAI) shipped Grok 4.6 just a month after 4.5 — a post-training upgrade rather than a new base model, with a 500K-token context window. It ties GPT-5.6 Sol at 61 on the Artificial Analysis Intelligence Index and overtakes Moonshot's Kimi K3, landing among the world's top-tier models, at a competitive $2 per million input and $6 per million output tokens. xAI attributes the gains to regenerated supervised fine-tuning trajectories and reinforcement learning in agentic environments.

Alongside the model, xAI launched Grok Bot in public beta: a team of agents that sign into apps, retain cross-task context, and share information among themselves, initially for SuperGrok Heavy and Cursor Ultra/Teams subscribers. That positions Grok squarely in the multi-agent orchestration race with OpenAI's agent tooling and Anthropic's Claude Code.

Third-party validation came quickly — Perplexity CEO Aravind Srinivas said his team benchmarked Grok 4.6 as an orchestrator on their Wide-And-Deep-Research benchmark and found it 'neatly sits on the Pareto frontier of performance vs cost,' now available to Pro and Max users. The pattern for xAI is rapid iteration on post-training rather than base-model leaps, betting that agentic RL and long context can keep pace with rivals' larger training runs. The open question is durability of these gains as OpenAI, Google and DeepSeek all push their own agent-tuned releases the same week.