Grok 4.7: coding, knowledge work and a faster serving option

SpaceXAI introduced Grok 4.7 for coding and professional knowledge work. Its announcement describes a larger base model, longer reinforcement learning and improvements to self-verification. Access includes Cursor, Grok Build and the Grok API. These are provider claims and availability statements; AZ Labs has not independently benchmarked the release.

The announcement gives starting rates of $2 per million input tokens and $6 per million output tokens. It also describes a fast variant with twice the output speed at twice the price. Check the rate that applies to your request before budgeting a long run. We do not infer an API identifier or context limit from the product name.

Our recommendation is to replay a small set of resolved engineering tasks. Include a bug with a reliable reproduction and a change that requires reading several modules. Compare accepted patches, time spent reviewing them and total billed usage. Give both models the same repository state and tool permissions.

A faster text stream is useful when a person is waiting for an answer. For an agent, the slowest part may be a build, search or external service. Measure the time until the task is complete before paying for faster output. Keep the existing deployment available during the evaluation so an unhelpful change is easy to reverse.