SpaceXAI Releases Grok 4.6: 500K Context, New xhigh Reasoning, Ties GPT-5.6 Sol Max

SpaceXAI introduced Grok 4.6 as a post-training upgrade of Grok 4.5, keeping the base model unchanged and improving through a longer supplemental training run, regenerated supervised fine-tuning trajectories, and reinforcement learning in agentic environments. The model scores 61 on the Artificial Analysis Intelligence Index, up from 56, tying with GPT-5.6 Sol Max, but trails on coding benchmarks such as DeepSWE (65.9% vs 73%) and Terminal-Bench (26%, though double Grok 4.5's 15.7%). Context length is 500,000 tokens with text and image input and text-only output, and it adds an xhigh reasoning-effort level. It is available via the xAI API as grok-4.6, in Grok Build, Cursor, OpenRouter, Vercel, and Cloudflare, but no open weights or self-hosting are offered. Pricing starts at $2 input / $0.50 cached input / $6 output per million tokens below 200K prompt tokens, doubling above that threshold. The article notes the bolded wins on GDPval-AA and AA-Briefcase are statistical ties within confidence intervals, and the comparison excludes Claude Opus 5.