"Key Takeaways"
- xAI launched Grok 4.7 on September 21, 2026 and calls it its most capable model for coding and knowledge work.
- xAI says it uses a larger base model than Grok 4.6, with a longer reinforcement-learning run aimed at harder, longer-running tasks.
- xAI also says the model is better at verifying its own work, managing longer context, and working inside the Grok Bot harness.
- API pricing starts at $2 per million input tokens and $6 per million output tokens. Treat benchmark and safety scores as xAI-reported.
Summarize this article with
Another Grok number does not help you write a better brief. The useful question is simpler: what feels different when you give it a real job?
Grok 4.7 is xAI's answer for coding and knowledge work. The company says the model works longer on hard tasks, checks itself more carefully, and ships at the same headline price and speed as Grok 4.6.
Grok 4.7 in plain English
Think of Grok 4.7 as the Grok AI model xAI wants you to use when the task is messy and takes time: a multi-file bug, a research pack, a document that has to survive review.
xAI positions it as its most capable model for coding and knowledge work. That is a company claim about its own lineup. Your own workload is still the real test.
What actually changed from Grok 4.6
According to xAI, three training changes matter more than a new marketing name. First, Grok 4.7 uses a larger base model than Grok 4.6. Second, it had a longer reinforcement-learning run on a harder mix of tasks, weighted toward problems that take many hours. Third, xAI says it trained the model to understand the Grok Bot harness natively, which it links to better conversational and knowledge-work behaviour.
xAI also says Grok 4.7 is better at verifying its own work and managing longer context. If that improvement holds up in real use, it should matter most on longer tasks: fewer wrong turns mid-task, and less forgetting what you already decided.
Why longer-running tasks matter
- Short prompts hide model weakness. Long jobs reveal whether the model keeps the original goal.
- Coding agents and research briefs often need hours of tool use, not one neat reply.
- If a model stalls halfway through a migration or a multi-file fix, you pay in babysitting time.
- xAI's own pitch centres on those longer jobs, not on another toy demo.
Self-verification is the interesting part
xAI says Grok 4.7 checks its own work more carefully. That does not mean the model is always right. It means the company is trying to reduce the number of times it finishes a task that still needs a full human rewrite.
Treat that as a reason to test, not as a guarantee.
Coding, knowledge work, and the charts
xAI reports stronger results than Grok 4.6 on several public-style benchmarks, including CursorBench 4.0, DeepSWE, EEBench, Terminal-Bench 4.0, and office-work suites such as AA Briefcase. On CursorBench 4.0, xAI reports Grok 4.7 at 46.3% against 40.4% for Grok 4.6 High. On Terminal-Bench 4.0, xAI reports 38.0% for Grok 4.7 against 20.3% for Grok 4.6 High.
Those are xAI-reported numbers on vendor-chosen comparisons. Other models still sit ahead on some of the same charts in xAI's own table. Use the scores as a map of where xAI wants attention, not as an independent ranking of every frontier model.
xAI also says Grok 4.7 improved on document and presentation work relative to Grok 4.6. If your week is slides and memos more than terminal sessions, that is the claim to check with your own files.
Price and availability, from xAI
| Detail | What xAI says |
|---|---|
| API input | From $2 per million tokens |
| API output | From $6 per million tokens |
| Same price class as | Grok 4.6 on the published comparison table |
| Where to use it | Grok, Cursor and Grok Build, the Grok API, third-party harnesses, and model routers or cloud platforms |
| Fast variant | Twice the output speed at twice the price, according to xAI |
Who should care
Try Grok 4.7 if you already live in Cursor or Grok Build, or if you pay for API coding agents and your jobs run long. The Grok 4.7 features that matter most for that group are stamina, self-checks, and Bot-harness awareness.
Wait if you only need short answers, light drafting, or a model you already trust for tone. A larger base model does not automatically make dinner plans better.
On safety, xAI says Grok 4.7 ships with a new safeguard stack and reports stronger refusal and jailbreak resistance on its own tests. Treat those as company-reported safety claims. They are not a reason to skip human review on anything that can move money or leak data.
Frequently Asked Questions
What is Grok 4.7?
It is xAI's September 21, 2026 model release for coding and knowledge work. xAI says it is stronger than Grok 4.6 on longer tasks and self-verification.
Is Grok 4.7 better than every other AI model?
No honest article can say that from a vendor post alone. xAI reports gains over Grok 4.6 and competitive scores against some frontier models. Your task is still the ranking that matters.
How much does the Grok AI model cost on the API?
xAI lists pricing starting at $2 per million input tokens and $6 per million output tokens, with a faster variant at twice the output speed and twice the price.
The Chai Takeaway
If your Grok or Cursor jobs die halfway through, give Grok 4.7 a fair trial on one real long task this week. If short chats already work, let the release notes pass. The useful upgrade is patience and self-checking, not a louder model name.
Explore More AI Updates