Grok 4.7 arrived on September 21 with a larger base model and more training on difficult coding and knowledge-work tasks. The release keeps Grok 4.6’s listed rates of $2 per million input tokens and $6 per million output tokens.
The developers say they used a longer reinforcement-learning run weighted toward problems that can take hours to complete. The model is intended to manage longer context and verify its own work more carefully.
On the published CursorBench 4.0 comparison, Grok 4.7 scores 46.3% against Grok 4.6’s 40.4%. The benchmark covers extended coding tasks, and the figures use the effort settings specified by the publisher. They do not establish which model will complete a particular project for the lowest total bill.
The announcement also reports improvements in document work and revised safeguards. Builders can access the model through the linked API route or try it in Grok.