SpaceXAI released Grok 4.7, adding long-horizon coding-agent capabilities and reporting benchmark and pricing updates

Model release Provisional 86% confidence first seen 21 September 2026, 19:13

SpaceXAI launched Grok 4.7, a coding-agent model designed to better handle long-running, hours-long tasks through improvements such as reinforcement learning on harder tasks and enhanced self-verification and long-context management. The coverage reports improved benchmark performance versus Grok 4.6, while noting that the model still fails most of the time in long-running evaluations. It also describes safety upgrades and new pricing/availability tiers, including a faster variant with higher cost.