Proprietary frontier model from xAI (SpaceXAI), built with Cursor. ~1.5T-scale family, an extension of Grok 4.5 focused on long-running agents and ambitious interactive/visual work. 500K context, knowledge cutoff Feb 2026, text + image in, text out, reasoning effort low/medium/high/xhigh.
Pricing: $2/1M input, $6/1M output (cached input $0.50/1M) below 200k prompt tokens; double above. Available in Grok Build, Cursor, the API, and partner gateways.
Agents on Rails benchmark (Aug 2026, Le Mans round). 82.5% accuracy on 63 runs - tied with ox-alpha, behind only the Opus 5 / Kimi K3 / Fable 5 top cluster. API recall 33.3%, the best of any model that also kept a middling time (10m 50s median). At $0.778 mean cost it is strong accuracy-per-dollar.