Grok 4 Features Explained: What Grok 4.7 Can Do, Cost and How to Use It

Last updated: October 7, 2026, 11:59 AM IST

xAI isn't slowing down. On September 21, 2026, Elon Musk's AI company released Grok 4.7 — its new frontier model aimed squarely at coding and knowledge work. It arrives with a 500,000-token context window, a brand-new reasoning tier, a faster variant, and benchmark scores that suggest the jump from 4.6 is bigger than the version number implies.

Whether you're a developer wondering if it's worth switching, or just tracking the Grok 4 features making headlines, here's everything that matters: what Grok 4.7 can do, what it costs, where you can use it, and how to try it today.

What Grok 4.7 Is

Grok 4.7 is xAI's latest flagship model, positioned as its best option for serious coding sessions and deep knowledge work — the kind of multi-hour tasks where weaker models lose the plot halfway through. It takes text and image input and produces text output, with a knowledge cutoff of February 1, 2026.

The headline upgrades over Grok 4.6 are speed and reasoning depth. There's a new "xhigh" reasoning effort tier above the previous high setting, and a Fast variant that delivers roughly twice the output speed for twice the price — built for when you're iterating fast and waiting on the model is the bottleneck.

Grok 4 Features: Context, Reasoning Tiers, and the Fast Variant

Let's unpack the specs that actually affect your work:

• 500,000-token context window: enough to hold entire codebases, long documents, or extended conversations in a single session without the model forgetting the beginning.
• Text + image input, text-only output: you can feed it screenshots, diagrams, and documents alongside your prompts.
• Reasoning effort tiers: low, medium, high (the default), and the new xhigh. Higher tiers think longer and harder — useful for tricky debugging and architecture decisions, overkill for simple questions.
• Fast variant: 2x output speed at 2x the price. Same model, less waiting.

The practical takeaway: xhigh reasoning plus the half-million-token window makes this a model built for long, complex sessions — exactly the "multi-hour coding" scenarios xAI is targeting.

The Benchmarks: Where It Actually Improved

Benchmarks aren't everything, but the differences here are hard to ignore — especially on agentic coding tasks:

• CursorBench 4.0: 46.3% — a solid showing on real-world coding assistance.
• DeepSWE v1.1: 71.0%, up from 65.2% on Grok 4.6 — a meaningful jump on software engineering tasks.
• Terminal-Bench 4.0: 38.0%, up from 20.3% on Grok 4.6 — nearly double, and the most striking improvement in the set.

That Terminal-Bench leap is the one to watch. It measures whether an AI agent can actually get things done in a terminal — installing packages, running tests, debugging failures — which is much closer to real developer work than answering quiz questions. For context on how fast this space moves, our Claude Sonnet 5.5 breakdown shows what the competition looks like right now.

Also Read
Claude Sonnet 5.5: What Is New, How Fast It Is, and What It Costs

Grok 4 Features and Cost: API Pricing Explained

For API users, Grok 4.7 keeps the same pricing as 4.6 — no increase, which is welcome:

• Input: $2 per 1 million tokens.
• Output: $6 per 1 million tokens.
• Cached input: $0.50 per 1 million tokens — a big saving if you're re-sending the same context.
• Above 200K prompt tokens, rates double to $4 input / $12 output — and note the catch: the entire request re-rates, not just the tokens past the threshold. Long-context work gets expensive fast, so keep an eye on prompt sizes.

The Fast variant doubles those numbers in exchange for roughly double the output speed. Whether that's worth it depends on how much you value your waiting time — for rapid iteration loops, many developers will happily pay it. TokensCost's Grok 4.7 pricing and benchmarks breakdown has the full rate table if you're budgeting API usage.

Where You Can Use It

xAI has pushed Grok 4.7 across just about every channel that matters:

• The Grok app — with Grok Build available to try for free, the easiest way to test-drive it.
• The xAI API — for developers building it into their own products.
• Amazon Bedrock, OpenRouter, Vercel, and Cloudflare — if you already work in those ecosystems, it's right there.
• Cursor — notably, Cursor's own model lineup now includes Grok 4.7, which tells you something about how seriously the coding-tool world is taking it.

That breadth of availability is a quiet advantage. You don't have to change your stack to try it — it's probably already an option wherever you work.

How to Try It Today

The fastest path depends on who you are:

• Curious individual: open the Grok app and try Grok Build free. Ask it something from your actual work — a bug you're stuck on, a document you need summarized — and judge it on your problem, not a demo.
• Developer: grab an API key from xAI or route through OpenRouter, and point a small project at it. Test the xhigh reasoning tier on something genuinely hard.
• Cursor user: select Grok 4.7 in your model picker and use it for a day of real coding. Compare it against whatever you normally use.

One tip: start with the default high reasoning tier before jumping to xhigh. You'll feel the speed difference, and for most tasks high is plenty. As Neowin's report on the launch notes, this release is really about sustained coding sessions — so test it on a long task, not a one-liner. And if you like tracking the model race, Reflection AI Beam's open-weight challenge shows the other side of the field.

Also Read
Reflection AI Beam: The Open-Weight AI Model Taking On DeepSeek and Qwen at 3x Less Compute

Frequently Asked Questions

When was Grok 4.7 released?

September 21, 2026, by xAI. It's the company's newest frontier model, focused on coding and knowledge work.

How much does Grok 4.7 cost?

Via API: $2 per million input tokens and $6 per million output tokens (same as Grok 4.6), with cached input at $0.50. Rates double above 200K prompt tokens. The Fast variant costs 2x for 2x the output speed.

What's the context window?

500,000 tokens — enough for entire codebases or very long documents in a single session. It accepts text and image input, with text-only output.

Where can I use Grok 4.7?

The Grok app (Grok Build is free to try), the xAI API, Amazon Bedrock, OpenRouter, Vercel, Cloudflare, and Cursor.

Is Grok 4.7 better than Grok 4.6?

On benchmarks, yes — notably DeepSWE v1.1 (71.0% vs 65.2%) and Terminal-Bench 4.0 (38.0% vs 20.3%). It also adds the xhigh reasoning tier and a Fast variant. Real-world gains depend on your workload, so test it yourself.

Sources

• Neowin — Grok 4.7 Brings Multi-Hour Coding and $2 Input Pricing (September 2026)
• TokensCost — Grok 4.7 Pricing, Context Window, Benchmarks

My Personal Take

My take: Grok 4.7 looks like xAI's most serious release yet, and the benchmark jumps on real coding tasks are hard to ignore — nearly doubling Terminal-Bench scores between generations doesn't happen by accident. The 500K context window plus the new xhigh reasoning tier tells me xAI is aiming squarely at developers grinding through long, messy coding sessions rather than casual chat users.

Should you switch? If you're already building with Grok or Cursor, trying 4.7 costs nothing and the speed gains alone might win you over. But I wouldn't abandon your current stack on benchmarks alone — run it on your own codebase for a week and see whether the reasoning holds up on your actual problems. Benchmarks are marketing; your own tests are the truth.