On September 21, 2026, xAI released Grok 4.7, the latest version of its flagship model. The company calls it its most capable model for coding and knowledge work.
Key improvements
According to the official announcement, Grok 4.7 is built on a new base model that is larger than Grok 4.6's. xAI also ran reinforcement learning for longer than before, on training data that includes many hard tasks that take hours to complete.
xAI says this led to the following improvements.
- Works on hard tasks for longer stretches
- Checks its own output more carefully
- Handles long context better
- Is better at creating documents and presentations
Key numbers
These are the main figures xAI published.
- 46.3% on CursorBench 4.0, a coding evaluation (Grok 4.6: 40.4%)
- 71.0% on DeepSWE v1.1, a software development evaluation (Grok 4.6: 65.2%)
- 19.6% on the Harvey Legal Agent Benchmark, a legal evaluation (Grok 4.6: 15.8%)
- 56.7% on HealthBench Professional, a clinical reasoning evaluation (Grok 4.6: 48.5%)
On safety, xAI says the model ships with rebuilt safeguards and is the strongest model it has tested at refusing inappropriate requests and resisting jailbreaks. All figures come from xAI's own announcement.
Pricing and availability
API pricing starts at $2 per million input tokens and $6 per million output tokens. A fast version with twice the output speed is also available, at twice the price of the standard version.
Grok 4.7 is available in Cursor, xAI's Grok Build developer tool and the Grok API, as well as through third-party coding tools, model routers and cloud services.
CoAI's take
This announcement is squarely aimed at developers and business use. Making the model available in Cursor on day one also suggests xAI is targeting adoption for coding. When comparing it with other major models, look beyond pricing and check the benchmark results closest to your own use case.
