Grok 4.7: What Changed and How to Try It for Free

Grok 4.7 is SpaceXAI's September 2026 model for longer coding and knowledge-work tasks. The company says it uses a larger base model than Grok 4.6, was trained on harder multi-hour tasks and checks its work more carefully.

It launched on September 21, 2026. This guide separates the company's documented claims from what you should verify in your own workflow.

What changed in Grok 4.7?

According to the official Grok 4.7 announcement, the model received a larger base model, a longer reinforcement-learning run and training weighted toward tasks that can take hours. SpaceXAI also says it improved long-context management and native understanding of the Grok Bot harness.

Those changes are aimed at work such as repository-scale coding, professional analysis and agent tasks that need intermediate checks. They do not establish that Grok 4.7 will beat another model on your application.

Launch claim	What to test yourself
Better long-task performance	Whether it preserves requirements across a realistic task
Stronger self-verification	Whether the final output actually passes independent checks
Longer-context management	Whether it finds and uses the right material, rather than merely accepting a large input
Better Grok Bot integration	Whether the harness exposes the controls and review flow you need

Vendor benchmarks help describe a target capability. They do not replace a controlled task from your own codebase or documents.

Where is Grok 4.7 available?

SpaceXAI lists Grok 4.7 in Cursor and Grok Build, through the Grok API, in third-party coding harnesses, and through model routers and cloud platforms. Each surface can add different tools, context rules and usage limits.

If you use Cursor, confirm that Grok 4.7 appears in the current model list for your plan. Selecting the same model in two products does not make their results identical: system instructions, repository indexing and tool permissions can change the outcome.

How to try Grok 4.7 for free

The launch page advertises free access through Grok Build. Treat this as an opportunity to evaluate the product, not a guarantee of unlimited ongoing usage.

Start with a small, inspectable project:

1. Open Grok Build and confirm that Grok 4.7 is selected.
2. Create a disposable project without private credentials.
3. Ask for one bounded feature with visible acceptance conditions.
4. Run the generated project and test the important path yourself.
5. Export or keep the work only after reviewing its dependencies and code.

Example task:

Build a small expense splitter with three inputs: participant names, expenses
and payer. Show each person's net balance. Validate empty names and negative
amounts. Include sample data and explain how I can test the calculation.

This task is small enough to verify, yet it tests UI, state and calculation logic. Do not begin with authentication, payments or sensitive production data.

API pricing and a safe first request

At launch, SpaceXAI listed Grok 4.7 from $2 per million input tokens and $6 per million output tokens. It also listed a faster variant at twice the output speed and twice the price. Pricing can change, so use the current console figures for budgeting.

Before connecting the API to an application, set a modest spending cap and test one short request. Record the model identifier, token usage, latency and response quality. Then test an invalid request and a timeout so the application does not treat every failure as a valid answer.

For agentic use, token price is only part of the cost. Repeated tool calls, long context and retries can dominate a session. Compare the cost of an accepted result rather than the advertised cost of one million tokens in isolation.

A practical coding evaluation

Use the same three-task framework you would apply to other AI coding tools: a reproducible bug, a multi-file refactor and a small feature. Begin from the same commit and specify the same tests.

Record:

• the product or harness running Grok 4.7;
• the exact model and settings;
• the prompt and starting commit;
• tests run and their output;
• manual interventions;
• review time and unnecessary changes.

If you compare Grok in Cursor with another model in a terminal agent, label it as a comparison of complete setups. The harness is part of what you are measuring.

Safety and verification

SpaceXAI says Grok 4.7 uses a new safeguard stack and reports improved performance on its safety and cybersecurity evaluations. These are vendor-reported results. They do not mean that generated code, professional advice or tool actions can skip review.

Keep write access narrow during evaluation. Do not give a new coding agent production credentials. For factual work, follow the AI answer verification checklist and open the sources it cites.

Who should try it?

Grok 4.7 is most interesting to developers already using Cursor or another supported harness, and to users who want a free Grok Build trial before considering API use. If your task is simple and your current model already performs acceptably, switching solely because a benchmark changed may add cost without improving the result.

Review this page after changes to Grok Build's free access, API pricing, model availability or Cursor support.