SpaceXAI launches Grok 4.7, its most powerful coding model
What's the story
Elon Musk's SpaceXAI (formerly known as xAI) has launched its latest and most powerful coding model, Grok 4.7. The new model is said to be twice as fast as its predecessor, but at half the price. According to SpaceXAI, the new version can work longer on difficult tasks, self-evaluate its progress, and develop more carefully.
Performance
Grok 4.7 has been tested using Cursor Bench 4.0
Grok 4.7 has been tested using the Cursor Bench 4.0 and has shown peak performance in long coding tasks. The model is said to be at the forefront of price and performance for such tasks.
It also features a more expansive base model than Grok 4.6, having undergone longer reinforcement learning training on a tougher mix of tasks weighted toward problems that take hours to complete.
Improvements
The new model is better at verifying its own work
The new model is better at verifying its own work and handling longer contexts.
It also natively understands the Grok Bot harness, making it better at conversational tasks and general knowledge work.
Grok Bot harness is SpaceXAI's agent orchestration and management layer that turns AI models into persistent digital coworkers.
Grok 4.7 has also been improved for document and presentation creation/curation, as per its performance in the GDP Val benchmark.
Security
Grok 4.7 comes with a new safeguard stack
Grok 4.7 comes with a new safeguard stack, making it one of the strongest models SpaceXAI has tested for refusal and jailbreak resistance.
In dual-use domains like cybersecurity and biological work, it strikes a balance between utility for benign tasks and safety.
The model is available now through Cursor and Grok Build at $2 per million input tokens and $6 per million output tokens.