Grok 4.7 Lands at the Same Price as Grok 4.6

Grok 4.7 is xAI's new coding-focused model, released on 21 September 2026 at exactly the price of the model it replaces: $2 per million input tokens and $6 per million output tokens. The headline figure is DeepSWE v1.1, where it posts 71.0% at high effort against Grok 4.6's 65.2%. If you already route work to 4.6, that is a capability increase you get without renegotiating anything. If you do not, the question worth asking is narrower than the launch post makes it sound.

What Grok 4.7 Changed

xAI describes 4.7 as a new, larger base model with extended training on demanding tasks, engineered to natively understand the Grok Bot harness. Coverage of the launch puts the base model at 2.1 trillion parameters against 1.5 trillion for its predecessor, roughly a 40% expansion.

The published scores, from xAI's own announcement:

The model is available in Cursor and Grok Build, and through the Grok API, third-party coding harnesses, model routers and cloud platforms. That is a wider day-one surface than the staged Bedrock rollout Grok 4.6 got.

The Price Is the Actual News

A 5.8 point jump on a software engineering benchmark is a good quarter's work. Shipping it at the previous model's price is the part that changes anyone's spreadsheet. Same input cost, same output cost, more capability per token spent.

There is a faster variant at double the output speed for double the price, $4 in and $12 out. That is a real decision rather than an obvious upgrade: you are paying 100% more to halve latency, which pays off for interactive coding and almost never pays off for batch work running overnight.

Where the Coding Win Does Not Extend

On CursorBench 4.0, which stresses longer-running tasks rather than single fixes, Grok 4.7 scores 46.3%. That beats GPT-5.6 Sol at 41.7% and its own predecessor at 40.4%, but Fable 5.1 leads the category at 51.8%. The same pattern shows on DeepSWE, where 71.0% sits close to Sol's 72.7% and slightly ahead of Fable 5.1's 70.0% without clearing the field.

So the accurate reading is that 4.7 is now competitive at the frontier for coding rather than ahead of it, and that its position depends on which coding benchmark you pick. That is normal, and it is why a single leaderboard position tells you less than it looks like it does.

These are also vendor-published numbers. They are useful as a direction of travel and weak as a purchasing decision. The only figure that settles anything is the one you generate on your own codebase.

What to Do About It

If you are on Grok 4.6, there is no pricing reason to wait. Re-run whatever evaluation set you used to choose 4.6 and see whether the gain shows up on your tasks rather than on SWE-style benchmarks.

If you are on a different provider, the honest answer is that a same-price upgrade at one vendor does not justify a migration. It justifies a test. Release weeks like this one are frequent enough now that reacting to each is a full-time job in itself, which is a good argument for a filter rather than a feed.

Sources: xAI, Introducing Grok 4.7 and MarkTechPost's launch coverage, both dated 21 September 2026.