xAI launched Grok 4.7 on September 21, a new model aimed primarily at coding and extended knowledge work. Unlike announcements that require waiting for an access list or a gradual rollout, this model is already available through an API and development tools.
Grok 4.7 is available through Grok API, the Grok Build environment, the Cursor development tool and other third-party systems that route requests to models.
Base pricing is 2 dollars per million input tokens and 6 dollars per million output tokens. xAI also offers a faster variant at twice the price.
This makes the launch particularly interesting for developers and organizations evaluating models on a combination of price, performance and availability rather than a benchmark headline alone.
What changed in Grok 4.7
According to xAI, the new model is larger than Grok 4.6 and underwent longer reinforcement learning training on a mix of complex tasks.
The company focuses the launch on two main areas: coding and knowledge work.
This is not just about writing short code snippets. xAI emphasizes longer tasks where the model needs to understand context, work through multiple stages and continue tackling a task over time.
AI companies are currently trying to differentiate their models in this area, especially for coding agents and tools working on large code repositories.
How much Grok 4.7 costs
Standard pricing is:
2 dollars per million input tokens and 6 dollars per million output tokens.
A faster version is also available at twice the price, which xAI says is designed to provide a higher output rate.
But token pricing is only part of the real cost.
A cheaper model can become expensive if it needs far more tokens, attempts or corrections to produce a useful result.
Teams evaluating Grok 4.7 should therefore measure cost per completed task, rather than just compare list prices per million tokens.
What the benchmarks say
xAI published a series of scores comparing Grok 4.7 with other models.
According to company data, Grok 4.7 scored 46.3% on CursorBench 4.0 and 71.0% on DeepSWE v1.1 in high effort mode.
These are interesting figures, but they were published by xAI rather than a broad independent evaluation.
Even in xAI's own tables, the model does not lead every test.
Other models score higher on some measures, so there is no basis to describe Grok 4.7 as a model that "beat every competitor".
The more interesting story is the relationship between xAI's reported performance and the price and availability.
What has not yet been published
Despite a relatively detailed launch page, several important details remain missing.
xAI did not publish an exact parameter count there, nor did it provide full details of the context window, rate limits or availability by country.
There is also not yet a substantial body of independent testing to establish how Grok 4.7 performs against other models in everyday work.
This matters especially in coding.
A benchmark can measure a particular capability, but a development team needs to know how the model works with its own code, libraries, standards and actual tasks.
What about safety
xAI also published safety and cybersecurity metrics and presents Grok 4.7 as having its most advanced safeguards to date.
These, too, are company data.
This does not mean the model is "hack-proof" or cannot be induced to perform problematic actions.
Safety metrics are part of the picture, but an organization connecting a model to code, development tools or internal systems still needs restricted permissions, testing and human approval for sensitive actions.
Who should test Grok 4.7
Developers already using Cursor or tools that make switching models easy can evaluate Grok 4.7 without a major workflow change.
For business teams, the question is less "who scored higher" and more whether the model delivers sufficient quality at a price suited to the workload.
The right way to test this is to build a small set of real tasks and evaluate Grok 4.7 against the model already in use.
Measure quality, time, the number of corrections and total cost.
For now, it is clear that Grok 4.7 is an available product rather than just an announcement. It is too early to determine whether xAI's performance and speed claims will hold up in independent testing.