xAI’s new public API model has a 500,000-token context window, text-and-image input and published prices of $2 per million input tokens and $6 per million output tokens. The faster serving tier costs twice as much and is currently limited to Cursor and Grok Build. Official documentation, updated September 21
The operational detail deserves more attention than the context number: xAI recommends a stable prompt_cache_key so successive turns reach the same server. Without it, an agent can repeatedly pay full input cost on cache-cold requests. For a real workflow, compare cost per completed task, including failed tool calls, repeated context and human repair. The vendor’s coding claims still need task-level evaluation; the launch at least supplies enough implementation detail to test them honestly.