01
Overview
Grok 4.7 is SpaceXAI's new frontier model for coding, agentic tasks and professional knowledge work. Released September 21, 2026, it succeeds Grok 4.6 with a larger base model, a longer reinforcement-learning run and training weighted more heavily toward difficult tasks that can take hours to complete.
02
What is Grok 4.7?
SpaceXAI positions Grok 4.7 as its most capable model for coding and knowledge work. The company says the model is better at checking its own work, managing long context and completing long-running tasks than Grok 4.6. Those performance statements are vendor claims; the benchmark figures below should be read as SpaceXAI-reported results unless an independent evaluation is explicitly identified.
The model accepts text and image input and produces text. The public API model ID is grok-4.7.
03
Grok 4.7 availability
Grok 4.7 launched on September 21, 2026 through the Grok API, Grok Build and Cursor, with SpaceXAI also saying it is available through third-party coding harnesses, model routers and cloud platforms.
SpaceXAI's API release notes document a U.S. regional endpoint. Availability through third-party providers can differ from direct xAI availability and should be checked with the relevant provider.
A separate Grok 4.7 Fast variant runs at roughly twice the output speed and twice the token price according to SpaceXAI. At launch, however, Fast is available through Cursor and Grok Build rather than the public xAI API.
04
Context window and modalities
Grok 4.7 has a 500,000-token context window. SpaceXAI documents text and image inputs with text output.
The model supports function calling and structured outputs, which makes it suitable for agentic applications that need to invoke external tools or return data conforming to a schema.
05
Reasoning levels
Developers can choose low, medium, high or xhigh reasoning effort. High is the documented default.
This matters for both latency and task complexity: a developer does not necessarily need the model's highest reasoning setting for every request. Benchmark comparisons also need to state the reasoning setting because SpaceXAI reports several headline Grok 4.7 results using xhigh.
06
Grok 4.7 API pricing
For direct xAI API usage below 200,000 prompt tokens, SpaceXAI lists:
- Input: $2 per million tokens
- Cached input: $0.50 per million tokens
- Output: $6 per million tokens
Requests exceeding 200,000 prompt tokens use higher-context pricing. SpaceXAI's September 21 release notes list $4 per million input tokens, $1 per million cached input tokens and $12 per million output tokens above that threshold.
Provider pricing is not necessarily identical. For example, OpenRouter can expose its own routed provider prices, so a third-party rate should not be presented as xAI's direct API price.
07
Rate limits
At launch, SpaceXAI's model documentation lists 150 requests per second and 50 million tokens per minute for the displayed U.S. deployment. Limits can be account-, region- or deployment-dependent and should be rechecked before relying on them for capacity planning.
08
What changed from Grok 4.6?
SpaceXAI says Grok 4.7 uses a new, larger base model and a longer reinforcement-learning run on a harder task mix. Training was weighted toward problems requiring multi-hour work. The company also says the model was trained to understand the Grok Bot harness natively and improved at self-verification and long-context management.
That makes the release more than a new API alias: it is a new model generation aimed particularly at long-running software engineering and knowledge-work agents.
09
Grok 4.7 benchmarks
SpaceXAI reports the following selected results for Grok 4.7:
- CursorBench 4.0: 46.3% at xhigh
- DeepSWE v1.1: 71.0% at high
- Terminal-Bench 4.0: 38.0% at xhigh
- Harvey Legal Agent Benchmark: 19.6% at xhigh
- HealthBench Professional: 56.7% at xhigh
- EEBench: 64.0% at xhigh
These are vendor-published results from the launch announcement. They should not be interpreted as independently reproduced measurements, and comparisons can depend heavily on harness, reasoning effort, prompts and evaluation setup.
SpaceXAI also reports improvements over Grok 4.6 on several of these tests, including CursorBench and Terminal-Bench.
10
Grok 4.7 for coding and agents
The combination of long context, configurable reasoning, function calling and structured outputs makes Grok 4.7 particularly relevant to coding agents and multi-step workflows. Grok Build is SpaceXAI's own coding-agent harness, while Cursor provides another launch surface.
Developers building their own systems can use the API rather than either interface. A separate implementation guide can cover request structure, reasoning settings, tool calling and long-context cost controls without bloating this overview.
11
Important limitations
Grok 4.7 is not an open-weight model. The public product is accessed through hosted services and APIs.
The 500K context window should not be confused with a promise that every application should routinely send 500K-token prompts. Direct xAI API pricing increases above 200K prompt tokens, so context management and caching can materially affect cost.
The Fast variant should also not be described as a public xAI API model unless SpaceXAI changes its availability; the September 21 documentation limits it to Cursor and Grok Build.
Finally, benchmark results from the launch are claims published by SpaceXAI. Independent evaluations may produce different results.
12
FAQ
When was Grok 4.7 released?
SpaceXAI released Grok 4.7 on September 21, 2026.
What is the Grok 4.7 context window?
The documented context window is 500,000 tokens.
Does Grok 4.7 support images?
Yes. SpaceXAI documents text and image input with text output.
How much does the Grok 4.7 API cost?
Direct xAI API pricing starts at $2 per million input tokens, $0.50 per million cached input tokens and $6 per million output tokens below 200K prompt tokens. Higher-context requests have higher rates.
Is Grok 4.7 on the API?
Yes. The public model ID is grok-4.7.
Is Grok 4.7 Fast on the public API?
Not at launch. SpaceXAI says the Fast variant is available through Cursor and Grok Build, not the public xAI API.
Does Grok 4.7 support tool calling?
Yes. SpaceXAI documents function calling and structured outputs.
13
Sources
Primary sources: SpaceXAI's September 21 Grok 4.7 announcement, model documentation and API release notes. OpenRouter is used only as a secondary source for third-party routing/availability context.
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.