The Grok 4 family
If you send long inputs, take Grok 4.3 rather than 4.5. The newer member of this family carries a 500,000 token window against 4.3 one million, charges $6 for output against $2.50, and the Batch API will not take it at all.
- 4.5 window is half of 4.3
- $6 output against $2.50
- batch discount only on the older models
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. What changed
The timeline, and an upgrade that takes three things back
The xAI catalogue prices these members today: Grok 4.5 with a 500,000 token window, Grok 4.3 with one million, and three 4.20 generation models that also carry one million, one of them multi-agent. Moving from 4.3 to 4.5 takes three things away from you and gives one back.
It takes: half the context window, the 20 percent batch discount that 4.3 has and 4.5 does not, and the ability to switch reasoning off. The reasoning guide states plainly that reasoning cannot be disabled on 4.5, while 4.3 accepts the same parameter with a value of none. It gives: a far better coding score, 1554 against 1355 on the WebDev Arena leaderboard, and much higher published rate limits.
The price went up rather than down: input from $1.25 to $2, output from $2.50 to $6. We saw the same direction in the GLM family and the opposite in the Opus family. So "newer means cheaper" is not a rule in this market, however often round-ups describe it that way.
grok-latest is not 4.5
The documentation says a bare model name is aliased to the latest stable version of that model. In the model data the same catalogue page publishes, grok-latest is listed among the aliases of Grok 4.3, not 4.5; the aliases on 4.5 are grok-4.5-latest and grok-build-latest.
The practical effect: anyone who wrote grok-latest in their code and believes they are on the flagship is running 4.3. Their invoice is smaller and so is their coding score. No product page says this. It only falls out of putting two rows of data next to each other.
Which one to take
For a coding agent and short prompts, 4.5. For anything with long inputs, or anything that runs as an overnight batch, 4.3: twice the window, under half the output price, and a 20 percent batch discount on top. For work that needs no reasoning at all and only wants a fast answer, the 4.20 generation even sells a separate non-reasoning model under its own id.
One constraint holds for all of them: none escapes the 200,000 token threshold. On every member of this family, a request whose prompt reaches that number bills every one of its tokens at double.
Release timeline
- Grok 4.5 current Half the context window of Grok 4.3, at more than double its output price.
Using it from Iran
This column carries a date and says how it was checked, because most listicles guess it. Where we have only read a vendor policy page, the note below says exactly that.
- Reachable
- not checked
- Payment
- not checked
- Free tier
- no
How we checked: x.ai returns 403 to our server on every path, so we have not read a page that would carry a country list. The developer documentation is open and carries none.
What it is good at
- Two genuine routes in one family: the pricier coding model and the cheaper long-input one
- Grok 4.3 at one million tokens and $2.50 output, the cheapest way into this family
- A 20 percent batch discount on 4.3 and on all three 4.20 generation models
- The 4.20 generation even ships a non-reasoning model and a multi-agent model under their own ids
Where it falls short
- The newest member has the smallest window: 500,000 tokens against one million.
- The line got more expensive: 4.5 output is more than double 4.3.
- The Batch API rejects Grok 4.5, so there is no volume discount on the flagship.
- The 200,000 token threshold doubles the whole request on every member, not just the tokens above it.
- No maximum output token figure is published for any member.
- Iran access is unchecked for all of them, because x.ai returns 403 to our server.
Our take
This is the only family in this reference where the flagship is the wrong choice for most work. If you feed it whole repositories or long documents, take 4.3. Reach for the flagship only when the job is code and the prompt stays short.
Questions people actually ask
Is Grok 4.5 better than Grok 4.3
For code yes, 1554 against 1355 on the web-dev leaderboard. For anything else it depends: 4.3 has twice the context window, charges under half for output, and takes a batch discount.
Which model is grok-latest
In the model data published by the xAI catalogue page, grok-latest sits among the aliases of Grok 4.3. If you want the flagship you have to name its id explicitly.
Which member works out cheapest
Grok 4.3 at $1.25 in and $2.50 out, with a further 20 percent off if your work runs as a batch. The cheaper catalogue model, Grok Build 0.1, is not part of this family and carries a 256,000 token window.
Sources
- xAI models vendor source 12 August 2026
- xAI pricing vendor source 12 August 2026
- xAI Batch API vendor source 12 August 2026
- xAI reasoning guide vendor source 12 August 2026