Grok 4.7: Same Price, Longer Coding Runs

xAI has released a frontier coding model that keeps list pricing flat and pushes harder on long, multi-hour tasks.

What happened

On 21/09/2026 SpaceXAI, the name xAI now trades under after the SpaceX acquisition, launched Grok 4.7.

The company positions it as its most capable model for coding and knowledge work, served at the same price and speed as Grok 4.6.

Day one surfaces include Cursor, Grok Build, the Grok API, third party coding harnesses, and model routers.

The public model id is grok-4.7.

Context stays in the 500,000 token class, with text and image input and text output.

List pricing under 200,000 prompt tokens is 2.00 US dollars per million input tokens, 0.50 US dollars for cached input, and 6.00 US dollars per million output tokens.

At current rates that is roughly £1.51, £0.38, and £4.52 respectively.

Once a prompt reaches 200,000 tokens, the whole request bills at double those rates.

A fast variant is mentioned at twice the output speed and twice the price, though the public docs page currently lists grok-4.7 as the main id.

xAI says the new model uses a larger base than 4.6 and a longer reinforcement learning run weighted toward problems that take many hours.

It is trained to check its own work more carefully, manage longer context, and understand the Grok Bot harness natively.

On CursorBench 4.0, which stresses longer running coding tasks, the launch table puts Grok 4.7 at 46.3% against 40.4% for Grok 4.6.

On Terminal-Bench 4.0 the same materials show a jump from 20.3% to about 38%.

DeepSWE v1.1 moves from 65.2% to 71.0% at high effort on the vendor card.

Those scores are self reported by xAI, so treat them as a buying signal to verify on your own repos rather than a finished proof.

Why this matters for UK businesses

Plenty of UK product and agency teams already pay for Cursor seats or call the Grok API for agent style coding.

A same price upgrade matters more than a flashy new sticker when budgets are fixed for the quarter.

You can swap the model id and keep the same forecast, then spend the week measuring whether longer runs finish with fewer retries.

That is useful for migration work, multi file refactors, and terminal heavy jobs where the old model ran out of patience.

Grok 4.7 is also stronger on the knowledge work benches xAI publishes for office style tasks, documents, and presentations.

For a British SME that means the same bill class can cover more of the brief writing, spreadsheet reasoning, and ticket triage that sits beside the code.

Availability inside Cursor matters for teams that do not want a separate chat window for every hard task.

Grok 4.6 remains available, and Cursor still positions Composer as the fast everyday coding model.

How we see it at Adevious AI

We would treat 4.7 as the new default for hard coding and agent loops, not as a blanket replace for every short chat.

Re measure token per task at high and xhigh effort.

Docs default reasoning effort to high, and several headline benches were reported at xhigh, so output can rise when the model stays in the job longer.

Watch the 200,000 token cliff.

A long prompt that tips over the threshold bills the entire request at the higher tier, which can surprise finance if you paste large repos by habit.

Do not hand unattended medical, legal, or safety critical decisions to the model.

xAI's own materials still assume human oversight in those domains, and absolute scores on legal and clinical benches remain well short of a solo operator.

If you already use Grok Bot style workflows, the native harness training is the part to trial first on real tickets rather than synthetic prompts.

What to do this week

If you use Cursor, select Grok 4.7 on one non critical project and compare merge ready diffs against your current default on the same brief.

If you call the API, change one staging route to grok-4.7 and log tokens, wall time, and human edit rate for a week.

Pin Grok 4.6 on any path where latency or short answers matter more than persistence.

Price your typical million token block in GBP before you raise effort levels, so the finance conversation stays boring.

Bottom line

Grok 4.7 is not a new price tier.

It is a same cost step toward longer coding and knowledge work runs, with day one access in Cursor and on the Grok API.

For UK teams already in the Grok stack, that is worth a controlled cutover this week rather than waiting for the next budget cycle.

Sources

SpaceXAI / xAI, Introducing Grok 4.7, 21/09/2026: https://x.ai/news/grok-4-7

xAI developer docs, Models and pricing for grok-4.7: https://docs.x.ai/developers/models

Cursor, Introducing Grok 4.7, 21/09/2026: https://cursor.com/blog/grok-4-7

Acronyms

API: Application Programming Interface

GBP: Great British Pound

SME: Small and Medium sized Enterprise

UK: United Kingdom

US: United States