The Grok API is xAI’s way of selling the Grok models to developers. Getting to a first useful result the direct way means an account, a key, billing, and code around the call. The short way is to pick Grok 4.6 inside Amelia’s Agent, where it costs $2 per million input tokens and $6 per million output tokens, prints its charge under each answer, and does not need you to write the loop around it. Try it on a job you already understand, read the price, and then decide whether to build on it.
Most people looking up the Grok API want to know two things: what it costs and whether it is any good for their work. Both have short answers, and neither requires signing up for anything.
The price, as we bill it
Inside Amelia’s Agent, Grok 4.6 is listed at $2.00 per million input tokens and $6.00 per million output tokens. That is the rate the platform charges you, printed on every answer down to the fraction of a cent, so there is no separate invoice to reconcile. It sits in the middle tier of the model list: pricier than the budget models that win most page builds, well under the frontier models that cost fifty times more per token.
A useful way to read that number: a one-page website build typically finishes in a single turn on a mid-tier model and costs a few cents. A scrape or a spreadsheet cleanup is usually a fraction of a cent. The price per token only becomes the story when a job takes many turns, and turns to done is what the model choice actually changes.
What it is good at
Grok 4.6’s profile in our catalogue is short and accurate: a strong general coder with quick turns, and it can see images, so it can look at a screenshot of what it just built. That makes it a sensible pick for ordinary web work, scripts, data wrangling and small apps, and a reasonable one for anything where a fast answer beats a deliberated one.
What we cannot tell you yet is how it ranks on our own benchmark, because most of our production runs have gone to the default model. We are honest about that in the Grok bot post, and we fixed the routing so harder jobs reach capable models regardless of how many samples they have. Grok 4.6 clears that capability floor.
The direct route, and what it costs in time
Going straight to the API means an xAI account, an API key, a payment method on file, and then the part nobody budgets for: the loop. A model call returns text. Turning that into a working thing means running the code, reading the error, feeding it back, checking the result, and stopping when the budget is spent. That is an afternoon of plumbing before you have learned anything about whether Grok suits your job.
The two-minute route
- Open the app, pick a job you already know the answer to. A scrape of a page you can check by eye is ideal.
- Choose Grok 4.6 from the model list, or leave Balanced on and see whether it is chosen.
- Read the price under the result and open the result itself.
That tells you, on a real task, how Grok 4.6 performs and what it costs, without a key or a billing page. If it suits you, keep choosing it. If it does not, the same job on another model is one click away, and the comparison is the whole point.
When to go direct anyway
If you are building your own product on top of Grok, you will eventually want your own key and your own loop, and the API is the right tool. Even then, the cheapest way to prototype is the one where the loop already exists: build the first version here, watch which model actually finishes your jobs, and take the code with you. It lands in a git repository you own.
What “Grok bot” actually means Which model for which job
Get the next one.
Working notes on building with AI, with the measurements and the bills attached. Sent when there is something worth sending, which is not weekly.
No spam, and one click to leave. We do not sell or share the list.
Grok 4.6, no key required.
Choose it from the model list. The price is on every answer, and the free trial covers the first jobs.
Free to start · no card · over 300 models, one account