Grok 4.7

What the vendor says. The Models page says that apart from dedicated audio, image and video models, developers should "use Grok 4.7" for everything else, "including code", and calls it the most capable model the company has built. The Grok 4.7 guide lists where it runs: the xAI API, a US regional endpoint, Grok Build (where it is the default model of the coding agent), Cursor on all plans, and several third-party AI gateways. The docs also note Grok has no knowledge of events after its training data unless server-side Web Search or X Search is enabled.

Our editorial judgment. The X Search tool is the clearest point of difference from the GPT-6 family: if your product needs answers grounded in recent X posts, Grok 4.7 is the obvious candidate to test. Its output price of USD 6.00 per 1M tokens is below GPT-6.1 Sol's USD 10.00, which matters for long generated answers. The lack of Batch support makes it a weaker fit for overnight bulk jobs. For coding assistants that bundle a model, see our AI coding tools guide; for consumer chat apps, see AI chatbots.

• Long-context pricing. Requests whose prompt reaches 200k tokens are billed at the higher rate for all tokens in the request: USD 4.00 input, USD 1.00 cached, USD 12.00 output per 1M tokens.
• US regional endpoint premium. Requests to https://us.api.x.ai/v1 keep inference in the United States and cost 1.1x: USD 2.20, USD 0.55 and USD 6.60 per 1M tokens below 200k, and USD 4.40, USD 1.10 and USD 13.20 above.
• No batch discount. The model page marks Batch API as not supported, and grok-4.7 is not in the pricing page's batch discount table.
• Tool fees on top of tokens. Web Search and Code Execution cost USD 5 per 1,000 calls; X Search costs USD 5 per 1,000 posts and USD 10 per 1,000 profiles returned.
• Fast variant is not on the API. Grok 4.7 Fast runs at 2x standard rates (1.5x for long context) but only in Cursor and Grok Build.
• Caching needs setup. The guide strongly recommends prompt_cache_key; without it, requests often pay full input price on a cache-cold server.
• Encrypted reasoning. The Responses API always returns reasoning.encrypted_content, which you pass back unchanged in multi-turn calls.
• Other limits. logprobs and top_logprobs are silently ignored; images must be jpg or png up to 20MiB.
• Rate limits. 150 requests per second and 50,000,000 tokens per minute, with higher limits by request to sales.

The Grok docs show earlier Grok models alongside 4.7. Grok 4.6 lists the same 500k context and the same prices; Grok 4.3 lists a 1M context at USD 1.25 input and USD 2.50 output and is eligible for a 20% batch discount. Against the OpenAI models we profile:

Grok 4.7 matches Sol on input price, undercuts it on output, and charges more for cached input (USD 0.50 versus USD 0.10). Sol and Astra offer a max reasoning level; Grok 4.7 tops out at xhigh.

1. Choose 30 or more real tasks, including some that need current information, and write a pass or fail rubric for each.
2. Run Grok 4.7 at its default high effort and at medium, alongside GPT-6.1 Sol at medium, with matched prompts and tools.
3. Set prompt_cache_key from the start so cached pricing reflects how you would run in production.
4. Track tokens, tool calls and prompt length per task, flagging any request that crosses 200k, and compute cost per accepted result.
5. If you need X data, test X Search separately and count posts returned, since billing is per item fetched, not per call.

We read the Grok 4.7 model page and its Markdown version, the Grok 4.7 guide, the Models page, the Pricing page and the x.ai homepage on 15 October 2026. We have not run Grok 4.7, and nothing here is a test result or ranking. Prices and limits change; confirm on the vendor page before budgeting. See our ranking method and the model directory for related profiles.

What is Grok 4.7? It is the Grok model the docs.x.ai pages describe as a frontier model for coding, agentic tasks and knowledge work, served through the Responses API.

How much does Grok 4.7 cost? It costs USD 2 per 1M input tokens and USD 6 per 1M output tokens below 200k prompt tokens at list price on 15 October 2026, doubling at or above 200k.

What is the Grok 4.7 context window? It is 500,000 tokens, and the guide states there is no text output limit.

Does Grok 4.7 support the Batch API? No, the model page lists Batch API as not supported.

Who makes Grok 4.7? The documentation at docs.x.ai and the x.ai homepage use the name SpaceXSI, and the x.ai footer copyright reads SpaceXAI LLC.

Can Grok 4.7 search X posts? Yes, X Search is a supported tool, billed at USD 5 per 1,000 posts and USD 10 per 1,000 profiles returned.