More from Sourceable
Continue reading our latest insights
Continue reading our latest insights
Grok 4.7 is an AI model from xAI designed for software development, complex tasks, and professional knowledge work. Announced on September 21, 2026, it focuses on longer coding tasks, better self-checking, and improved handling of complex workflows.
According to xAI’s official announcement, the model uses a larger base model than Grok 4.6 and underwent additional reinforcement learning. The goal is to help it work through difficult tasks for longer and check its results more carefully.
For developers, this matters because real software projects require more than generating code. A useful coding model must understand existing files, find bugs, make changes, and verify whether those changes work.
Grok 4.7 focuses on three key improvements.
1. Better performance on complex coding tasks
Grok 4.7 was trained with more emphasis on tasks that take hours rather than minutes. This makes it relevant to repository-level coding, debugging, and multi-step development workflows.
However, benchmark performance does not guarantee that every real-world coding task will be easier.
2. Improved self-checking
xAI says Grok 4.7 is better at verifying its own work. This can help when a coding agent needs to review a change, identify an error, or refine an earlier answer.
Developers should still run tests, review generated code, and check security-sensitive changes themselves.
3. Better support for long workflows
The model is designed to manage longer context and work with the Grok Bot harness. It is intended for coding and broader knowledge work, including tasks that require several steps.
The Grok 4.7 developer documentation lists a 500,000-token context window, image input, and reasoning settings from low to extra high.
Benchmarks offer a useful way to compare AI models. The results below come from xAI’s published launch comparison, not an independent test conducted for this article.
BenchmarkGrok 4.7Grok 4.6GPT-5.6 SolFable 5.1CursorBench 4.046.3%40.4%41.7%51.8%DeepSWE v1.171.0%*65.2%72.7%70.0%
The Grok 4.7 DeepSWE result was reported at high reasoning effort. The models were tested under different effort settings, so the figures are not a perfectly equal comparison.
Grok 4.7 improves on its predecessor. Its CursorBench 4.0 score rises from 40.4% to 46.3%. Its DeepSWE score also improves from 65.2% to 71.0%.
Grok 4.7 is competitive, but it does not lead every test. It scores above GPT-5.6 Sol on CursorBench, but below it on DeepSWE. Fable 5.1 also scores higher on CursorBench.
The key lesson is simple: one benchmark cannot tell you which AI coding model is best for every project. Results depend on the task, reasoning effort, tools, and testing setup.
Developers should compare AI models based on their daily work, not just a leaderboard.
Grok 4.7: Worth testing for longer coding tasks, debugging, and cost-conscious development workflows.
GPT-5.6 Sol: Scores higher on the DeepSWE result shown above, though it scores lower on CursorBench in xAI’s comparison.
Fable 5.1: Scores higher on CursorBench, making it a strong competitor in that particular test.
These are benchmark-specific observations, not a universal ranking. Test the models on your own codebase before choosing one for production.
When evaluating an AI coding model, check whether it can follow your project conventions, edit multiple files correctly, pass tests, and explain its changes clearly.
xAI’s developer documentation lists the following standard API rates:
Pricing item | Published rate |
|---|---|
Input tokens | $2 per million |
Output tokens | $6 per million |
Context window | 500,000 tokens |
Model ID |
|
These rates are published in the Grok 4.7 developer guide. Check the official pricing page before using them in a budget, since rates and billing conditions can change.
Grok 4.7 is available through the xAI API, Grok Build, Cursor, and supported third-party integrations. Availability depends on the platform.
The documentation also describes Grok 4.7 Fast, a faster variant with different pricing. Check the current terms if speed is more important than token cost.
Grok 4.7 may be useful for several common development tasks:
Debugging: Find likely causes of errors and suggest fixes.
Code review: Identify possible logic issues and missing edge cases.
Repository work: Make changes across related files.
Test creation: Draft unit tests and suggest additional cases.
Technical documentation: Explain code and prepare project documentation.
Start with a small task. Ask the model to explain its approach, make the change, and run the relevant tests. Then compare the result with another model if the task is complex or business-critical.
Before adopting Grok 4.7, test it against your real development needs.
Code quality: Does the solution work and follow your standards?
Test success: Does it pass existing tests without breaking other features?
Cost: How many input and output tokens does a task consume?
Speed: How long does it take to finish a complete task?
Reliability: Does it handle repeated runs consistently?
Security: Does it avoid unsafe code and protect sensitive information?
Record the results across several tasks. This gives you a more useful comparison than relying on one public benchmark.
Is Grok 4.7 good for coding?
Grok 4.7 shows improvements over Grok 4.6 on the coding benchmarks published by xAI. Its real-world value depends on your programming language, tools, and task complexity.
What is Grok 4.7’s coding benchmark score?
In xAI’s published comparison, Grok 4.7 scores 46.3% on CursorBench 4.0 and 71.0% on DeepSWE v1.1 at high reasoning effort.
How much does Grok 4.7 cost?
The published standard API rates are $2 per million input tokens and $6 per million output tokens. Check xAI’s developer documentation for current pricing and conditions.
Is Grok 4.7 better than GPT?
Not in every test. Grok 4.7 scores higher on CursorBench in xAI’s comparison, while GPT-5.6 Sol scores higher on DeepSWE. Choose based on your own tests and workload.
Grok 4.7 is a notable update for developers who need an AI model for longer coding tasks and complex workflows. Its improved scores over Grok 4.6 and published API pricing make it worth testing.
Still, it does not win every benchmark. The best choice depends on code quality, task success, speed, and total cost. Test Grok 4.7 on your own projects before making it a core part of your development process.