xAI releases Grok 4.7: a larger base model built for coding and long agent runs
A 500k-token context window, a new and larger base model, and a longer reinforcement-learning run aimed specifically at multi-hour agentic tasks and self-verification -- at the same token pricing as Grok 4.6.
xAI released Grok 4.7 on September 21, positioned as its latest frontier model for coding, agentic work, and professional knowledge tasks. The model carries a 500k-token context window with text and image input, flexible reasoning levels, and is live in Cursor, Grok Build, the Grok API, and through third-party routers and harnesses.
What actually changed under the hood
xAI describes Grok 4.7 as using a new, larger base model plus a longer reinforcement-learning run that weights training toward difficult, hours-long tasks -- a training-time investment aimed squarely at agentic workloads rather than short single-turn prompts. xAI also claims improved self-verification and long-context management, which would matter most on exactly the kind of extended agent runs the training was tuned for; these are xAI's own characterizations and haven't been independently benchmarked yet.
Pricing holds flat
Token pricing is unchanged from Grok 4.6: $2.00 per million input tokens ($0.50 cached) and $6.00 per million output tokens for prompts under 200k tokens, rising to $4.00/$1.00/$12.00 above that threshold. Holding pricing flat while shipping a larger base model is a different move than OpenAI's and Anthropic's price cuts the same week -- xAI is spending the efficiency gain on capability instead of passing it through as a lower bill.