Quick Answer
Grok 4.7 is a new SpaceXAI model for coding and knowledge work, with a 500,000-token maximum context window and configurable reasoning effort. SpaceXAI released it September 21, 2026, with standard API pricing starting at $2 per million input tokens. Developers should compare the context, reasoning setting, access route, and token costs before moving production workloads.
Key Takeaways
- Grok 4.7 has a 500,000-token maximum context window.
- The standard API costs $2 per million input tokens and $6 per million output tokens below 200,000 tokens.
- Grok 4.7 offers low, medium, high, and xhigh reasoning-effort settings.
- SpaceXAI reported a 46.3% CursorBench 4.0 score, a vendor-reported result.
- Grok 4.7 Fast costs twice the standard rates and is not available through the public API.
What is Grok 4.7?
Grok 4.7 is SpaceXAI’s newest model for coding and knowledge work. SpaceXAI released Grok 4.7 on September 21, 2026, and describes the model as its most capable release for those tasks, using a larger base model than Grok 4.6 and longer reinforcement-learning training on harder, multi-hour tasks. SpaceXAI’s launch announcement describes the release and its stated training approach.
Grok 4.7 matters primarily because the model is designed for work that can require a large amount of source material, multiple files, and extended reasoning. A larger model and longer training process do not guarantee better results for every prompt, however. Developers should validate outputs against their own codebase, documentation, and test suite before relying on Grok 4.7 for production decisions.
Grok 4.7 enters a market where coding tools increasingly connect models to files, search tools, and command-line workflows. Developers comparing model-assisted development options can also consider how parallel coding workflows affect review practices, permissions, and the amount of work a single AI session can delegate.
What does Grok 4.7’s 500,000-token context window mean?
Grok 4.7 supports a 500,000-token maximum context window, according to SpaceXAI’s model documentation. The model accepts text and image inputs and produces text output, which gives developers room to provide long documents, code repositories, images, and prior conversation material in one request. SpaceXAI’s Grok 4.7 documentation lists the context limit and supported input and output formats.
A context window is the amount of material a model can consider during a request. A 500,000-token maximum does not mean every use case needs or benefits from that much information. Larger prompts can cost more, take more planning, and still include irrelevant material that distracts from the task. The practical approach is to provide the files and instructions that directly affect the requested work.
Grok 4.7’s long context may be useful for reviewing a large collection of source files or comparing lengthy technical documents. Developers should still separate sensitive credentials, customer data, and proprietary information from prompts unless their organization’s AI policy and the selected platform’s controls explicitly permit that use.
How much does Grok 4.7 cost through the API?
Grok 4.7 costs $2 per million input tokens, $0.50 per million cached input tokens, and $6 per million output tokens for prompts under 200,000 tokens. Prompts above 200,000 tokens cost $4 per million input tokens, $1 per million cached input tokens, and $12 per million output tokens. SpaceXAI lists these standard API rates in its model documentation, so teams should calculate costs around both prompt length and the amount of generated output.
| Grok 4.7 API usage | Input tokens | Cached input tokens | Output tokens |
|---|---|---|---|
| Prompts under 200,000 tokens | $2 per million | $0.50 per million | $6 per million |
| Prompts above 200,000 tokens | $4 per million | $1 per million | $12 per million |
Grok 4.7 pricing changes at the 200,000-token threshold, which matters for teams using large repository snapshots, long transcripts, or extensive reference materials. Cached input is less expensive than standard input under both pricing tiers. The sensible planning method is to monitor token use by task type before expanding access across a team.
Grok 4.7 Fast costs twice the standard token rates, and SpaceXAI limits that variant to Cursor and Grok Build rather than the public API. Organizations that need a particular access route should confirm both model availability and billing terms before changing an existing AI development workflow.
Which reasoning controls does Grok 4.7 offer?
Grok 4.7 offers 4 reasoning-effort settings: low, medium, high, and xhigh. High is the default setting, according to SpaceXAI’s documentation. The controls give developers a way to select the model’s reasoning effort for different requests instead of treating every task as equally complex.
Grok 4.7 reasoning controls matter because a short code explanation and a difficult multi-file debugging task do not necessarily require the same level of model effort. Higher reasoning settings can be appropriate for more complex work, while lower settings may fit simpler tasks. SpaceXAI does not present the controls as a replacement for code review, testing, or human approval.
Developers should test the available settings against a defined set of internal tasks, such as code explanation, bug triage, refactoring, and documentation drafting. A useful evaluation compares accuracy, response quality, token consumption, and the amount of correction required after the model responds.
Where can developers access Grok 4.7?
Grok 4.7 is available in Cursor, Grok Build, the Grok API, third-party coding harnesses, model routers, and cloud platforms. SpaceXAI announced those access routes on September 21, 2026, which gives developers several ways to evaluate the model without assuming that every feature appears in every integration.
Cursor lists a 256,000-token standard context window for Grok 4.7 and a 500,000-token long-context mode. Cursor also lists tool access that includes web search, file search, editing, and shell commands. Cursor’s Grok 4.7 documentation sets out those integration details, including the different context options.
Grok 4.7 access through a coding environment can introduce additional permissions beyond the model itself. File search, editing, and shell commands can be useful in a development workflow, but teams should review repository access, command approval settings, and data-handling policies before enabling tools. The same caution applies to AI agents that can act across connected services, including emerging AI agent controls for smart-home platforms.
How did Grok 4.7 perform on CursorBench 4.0?
Grok 4.7 scored 46.3% on CursorBench 4.0 in SpaceXAI’s reported results, compared with 40.4% for Grok 4.6 High. SpaceXAI published those figures as part of the Grok 4.7 launch, and the results are vendor-reported benchmark scores rather than independent testing. The reported difference suggests an improvement under that benchmark’s conditions, but it does not establish how every coding team will experience the model.
Grok 4.7 benchmark results matter because coding models are often evaluated with task suites that do not fully match a company’s language choices, test requirements, security practices, or repository structure. A benchmark can help identify a model worth evaluating, but it cannot replace an internal trial using representative work.
Developers should measure Grok 4.7 against the tasks that affect their own release process. A careful evaluation includes correct code generation, regression rates, the quality of explanations, tool permissions, and whether reviewers can understand and maintain the resulting changes. Model claims should remain separate from verified results in a specific production environment.
Who should consider using Grok 4.7?
Grok 4.7 is most relevant to developers and knowledge workers who need long-context analysis, configurable reasoning effort, or access through supported coding platforms. The 500,000-token maximum context window and multiple access routes give the model a clear fit for teams evaluating AI assistance across substantial technical materials.
Grok 4.7 may be less suitable for teams that only need short, low-cost prompts or that cannot send work materials to an external AI service under their security rules. API costs also rise for prompts above 200,000 tokens, so the largest context option is not automatically the most economical option. Teams should define what data can be shared and identify the approval process before a pilot begins.
AI-assisted coding also requires routine account and extension security. Malware campaigns have used familiar AI and cryptocurrency themes to target users, including attacks involving malicious browser extensions. Developers should obtain tools from official sources, review requested permissions, and keep credentials out of prompts and project files.
What should developers test before adopting Grok 4.7?
Developers should test Grok 4.7 with a limited set of representative tasks before adopting it for broad production use. The first tests should cover the actual work the team expects to assign, such as explaining unfamiliar code, proposing a refactor, analyzing documentation, or locating a likely defect across multiple files.
- Define 3 to 5 representative coding or knowledge-work tasks with known evaluation criteria.
- Choose the reasoning-effort setting for each task and record the selected setting.
- Measure token use, output quality, correction time, and test results for each response.
- Review what files, tools, and commands the selected integration can access.
- Expand access only after the team has documented acceptable data use and human review requirements.
Grok 4.7 evaluations should include a stop line for sensitive data and high-impact actions. Stop and consult the security team or platform administrator before connecting the model to production repositories, enabling shell commands with broad permissions, or submitting regulated customer information. The practical goal is to test model usefulness without giving an AI integration more access than the task requires.
FAQ
Is Grok 4.7 available through the public API?
Yes, Grok 4.7 is available through the Grok API. SpaceXAI also lists Cursor, Grok Build, third-party coding harnesses, model routers, and cloud platforms as access routes.
What is the maximum context window for Grok 4.7?
Grok 4.7 has a 500,000-token maximum context window. Cursor lists a 256,000-token standard context window and a 500,000-token long-context mode for its integration.
How much does Grok 4.7 cost?
Grok 4.7 standard API pricing starts at $2 per million input tokens for prompts under 200,000 tokens. Output tokens cost $6 per million under that threshold, while prompts above 200,000 tokens use higher rates.
Does Grok 4.7 have configurable reasoning settings?
Yes, Grok 4.7 supports low, medium, high, and xhigh reasoning-effort settings. High is the default setting listed in SpaceXAI’s documentation.
Is Grok 4.7 Fast available through the public API?
No, Grok 4.7 Fast is not available through the public API. SpaceXAI says the faster variant is available only in Cursor and Grok Build and costs twice the standard token rates.