SpaceXAI Releases Grok 4.6: A 500K-Context Frontier Model Tuned for Long-Running Agents, Coding, and Knowledge Work

Details

xAI released Grok 4.6 on August 12, 2026, as a post-training upgrade to Grok 4.5 rather than a new base model. The company kept the foundation unchanged and invested the improvement in a longer supplemental training run using curated model-generated data, regenerated supervised fine-tuning trajectories, and reinforcement learning in agentic environments including knowledge work, general coding, web development, CAD design, and GPU kernel optimization. The model scores 61 on the Artificial Analysis Intelligence Index, up five points from Grok 4.5 and tied with GPT-5.6 Sol Max. Grok 4.6 extends context to 500,000 tokens and adds a new xhigh reasoning-effort level above Grok 4.5's highest tier. Pricing is $2 input / $6 output per million tokens via the xAI API. Deployment is live today. The model is available as grok-4.6 on the xAI API, is the default in Grok Build, ships in Cursor on all plans, and is routable via OpenRouter, Vercel, and Cloudflare. There is no open-weights release or self-hosting path. xAI reports increased self-testing and verification behavior on longer trajectories — a behavioral improvement from internal testing rather than independently verified benchmarks. The strongest fit is mid-market engineering teams running repository-wide refactors, 500K-token research-and-synthesis pipelines, or GPU kernel optimization. Regulated enterprises should treat this as a pilot candidate given procurement sensitivities around the xAI brand.