Grok 4.7 Review

Grok 4.7 is the frontier model from SpaceXSI (SpaceX Super Intelligence LLC, the company behind Grok, formerly branded xAI), announced on September 21, 2026. SpaceXSI calls it its most capable model for coding and knowledge work: it uses a new, larger base model than Grok 4.6, was trained with a longer reinforcement-learning run weighted toward multi-hour tasks, and is described as better at checking its own work, managing long context and producing documents and presentations. It ships with a new safeguard stack, which SpaceXSI says gives the strongest refusal and jailbreak resistance it has tested.

On the API the model name is grok-4.7. It takes text and images and returns text, with a 500,000-token context window, function calling, structured outputs, Batch API support and configurable reasoning effort (low, medium, high by default, and xhigh). Its knowledge cut-off is May 2026; live information needs the Web Search or X Search tools. Grok 4.7 is also available in Cursor and Grok Build, SpaceXSI's coding agent, plus third-party coding harnesses, model routers and cloud platforms. A Grok 4.7 Fast variant (same model, about twice the output speed at twice the price) is offered only in Cursor and Grok Build.

Worked examples at standard rates: a request with 100,000 input tokens and 10,000 output tokens costs 0.1 x $2 + 0.01 x $6 = $0.26. A 250,000-token prompt with 10,000 output tokens crosses the 200k line, so it costs 0.25 x $4 + 0.01 x $12 = $1.12. Reasoning tokens are billed as output.

Server-side tools are extra: Web Search and Code Execution cost $5 per 1,000 calls, X Search $5 per 1,000 posts returned, and Collections Search $2.50 per 1,000 calls. Rate limits for grok-4.7 are 150 requests per second and 50 million tokens per minute.

SpaceXSI says Grok 4.7 can be tried for free in Grok Build; consumer Grok and SuperGrok subscription prices were not readable on its pricing page and are not listed here.

Prices checked on the vendor's pricing page on November 10, 2026. See SpaceXSI API pricing and the Grok 4.7 model page.

What SpaceXSI's announcement and API docs list:

Grok 4.7 suits developers who want a frontier-class coding and agent model at a lower list price than most rivals: SpaceXSI's own table lists GPT-5.6 Sol at $4 / $20 and Fable 5.1 at $10 / $50 per million tokens against Grok 4.7's $2 / $6. It fits long agent loops in Cursor or Grok Build and API workloads that benefit from a 500k context. Watch the bill on high effort: reviewers found most output tokens go to hidden reasoning, so low effort can be cheaper for short, well-specified tasks.

If you want the strongest scores on long terminal tasks, SpaceXSI's own benchmarks show Claude Fable 5 ahead. Compare GPT-5.6 or GPT-6 Astra for OpenAI's line-up, and Grok 4.6 if you already run the previous model at the same price.

Early reviews call Grok 4.7 a clear step up from 4.6 that still trails the very top models. ComputingForGeeks ran nine infrastructure tasks on the API: Terraform and bash outputs passed, but Kubernetes manifests that passed linting crash-looped until the error was fed back; it found about 87% of high-effort output tokens were hidden reasoning, and low effort gave equally passing results at roughly 40% of the cost. In the Cursor forum thread, some users find it easier to talk to than 4.6, while others report nearly twice the tokens per task, a more verbose style, and object that launch charts compared 4.7 at xhigh with 4.6 at high. Most published benchmark numbers are SpaceXSI's own.