01
Overview
Grok 4.7 is available through SpaceXAI's public API as grok-4.7. This guide focuses on the implementation questions that are distinct from the broader Grok 4.7 overview: which model ID to use, how reasoning effort works, what tool and structured-output capabilities are documented, how the 500K context window affects cost, and which launch limits developers should verify before production use.
02
Before you start
You need an xAI API account and API key. Use SpaceXAI's current developer documentation for authentication and endpoint details rather than copying keys into source code.
The launch model ID is:
grok-4.7
Do not use grok-4.7-build-fast as though it were a public API model. SpaceXAI's September 21 release notes say the Fast variant is available in Cursor and Grok Build, not through the public xAI API.
03
Supported inputs and outputs
SpaceXAI documents Grok 4.7 as accepting text and image input and producing text output. The model has a 500,000-token context window.
It supports function calling, structured outputs and configurable reasoning. These features are especially useful when Grok is being used as part of an agent rather than as a simple chat completion.
04
Choose a reasoning effort
Grok 4.7 supports four documented reasoning settings:
- low
- medium
- high
- xhigh
High is the default.
Use reasoning effort as a workload control rather than automatically choosing xhigh. Straightforward extraction, classification or tool-routing tasks may not need the same reasoning budget as complex debugging or multi-hour engineering work.
SpaceXAI's headline launch benchmarks often use xhigh, so do not assume that the same results describe lower-effort production calls.
05
Function calling
Function calling lets Grok return a request to invoke an external tool or application function instead of pretending it has performed an action itself.
A production agent generally needs three layers:
1. Your application defines the tools and their input schemas.
2. Grok decides when a tool should be called and returns the structured tool request.
3. Your application validates and executes the call, then returns the result to the model when another model turn is required.
Treat tool execution as an application-security boundary. The fact that a model can request a function does not mean it should have unrestricted permission to execute destructive operations.
06
Structured outputs
Grok 4.7 supports structured outputs. This is useful when an application needs machine-readable data conforming to a defined schema rather than free-form prose.
Typical uses include extracting fields from documents, producing normalized research records, creating agent plans consumed by another service, or returning predictable application state.
Schema validation should still happen in your application. Structured generation reduces formatting uncertainty; it does not remove the need to validate values or business rules.
07
Working with the 500K context window
The documented context window is 500,000 tokens, but sending the maximum possible context is not automatically the best strategy.
SpaceXAI applies higher direct-API token rates when prompt length exceeds 200,000 tokens. Below that threshold, launch pricing is $2 per million input tokens, $0.50 per million cached input tokens and $6 per million output tokens. Above 200K prompt tokens, the September 21 release notes list $4 input, $1 cached input and $12 output per million tokens.
That makes context design a cost decision. Retrieval, summarization, prompt caching and removing irrelevant history can be useful even when the model technically has room for the full context.
08
Prompt caching
SpaceXAI lists cached input at a lower token rate than ordinary input. For applications that repeatedly reuse a large stable prefix, caching can therefore change the economics materially.
Do not assume every token in a repeated prompt will automatically qualify as cached input. Follow the current xAI caching documentation and inspect actual API usage/billing for the workload.
09
Rate limits
At launch, the Grok 4.7 model page displays 150 requests per second and 50 million tokens per minute for the shown U.S. deployment.
Treat those figures as current documentation, not universal guarantees. Limits can depend on account, deployment, region or later policy changes. Production systems should handle rate-limit responses and recheck their account's current limits.
10
Regional endpoint
SpaceXAI's September 21 release notes say Grok 4.7 is also served on its U.S. regional endpoint. Developers with residency or architecture requirements should verify the current regional documentation rather than inferring that every xAI region supports the model.
11
Grok 4.7 in Grok Build and Cursor
The same model also launched through Grok Build and Cursor. Those products are useful when the goal is an interactive coding-agent experience, while the API is appropriate when Grok needs to be embedded into your own application, workflow or agent.
Grok Build also supports extensions such as MCP servers, plugins, hooks and subagents, but those are capabilities of the Grok Build harness rather than intrinsic API capabilities of the Grok 4.7 model. Keep the model and the agent harness conceptually separate.
12
Direct xAI API vs model routers
Grok 4.7 is also appearing through third-party routers and platforms. Their pricing, rate limits, routing behavior and availability can differ from the direct xAI API.
If a router advertises a different token rate, describe it as that provider's price rather than replacing xAI's official direct price. This distinction is especially important during the first days after a model launch, when providers can change rates or add promotional pricing quickly.
13
Production checklist
Before deploying Grok 4.7 in a production agent:
- confirm the current grok-4.7 model documentation
- select the lowest reasoning effort that reliably handles the task
- validate every function/tool call in application code
- use structured outputs when downstream systems need predictable schemas
- control long-context usage rather than filling 500K tokens by default
- account for the higher xAI rates above 200K prompt tokens
- check caching behavior using actual usage data
- implement rate-limit and transient-error handling
- verify regional requirements
- recheck current pricing and limits before forecasting spend
14
FAQ
What model ID should I use for Grok 4.7?
SpaceXAI documents grok-4.7 for the public API.
Does Grok 4.7 support function calling?
Yes. Function calling is listed in the official model capabilities.
Does Grok 4.7 support structured output?
Yes. SpaceXAI documents structured outputs for the model.
What reasoning levels does Grok 4.7 support?
Low, medium, high and xhigh. High is the documented default.
Can I send 500K tokens?
The documented context window is 500,000 tokens. However, direct xAI pricing rises when prompt tokens exceed 200K, so large-context usage should be designed deliberately.
Is Grok 4.7 Fast available through the xAI API?
Not according to the September 21 launch documentation. Fast is listed for Cursor and Grok Build rather than the public API.
15
Sources
This guide uses SpaceXAI's official Grok 4.7 model documentation, September 21 API release notes and launch announcement. Recheck all limits and prices before production deployment because launch-era API details can change quickly.
Sources
Primary and supporting sources
Use the linked documentation to check current pricing, access and product behaviour. The dates above describe this article; they do not guarantee that every vendor claim remains current.