Grok 4.6 costs $2 per million input tokens and $6 per million output tokens on prompts under 200,000 tokens, with cached input at $0.50, read from the xAI model documentation on 16 August 2026. Past 200,000 tokens every line doubles: $4 input, $1 cached, $12 output.
That threshold, not the headline rate, is what decides a Grok bill, and it is the one thing most comparison tables leave out.
What grok-4.6 costs, on both grids
xAI publishes two prices for the same model, and which one applies depends on the length of the prompt you send, not on your monthly volume or your plan.
| Billing line | Under 200k tokens | 200k and over |
|---|---|---|
| Input | $2.00 | $4.00 |
| Cached input | $0.50 | $1.00 |
| Output | $6.00 | $12.00 |
grok-4.6 at $2 in and $6 out per million under 200k, read 16 August 2026
The consequence is unusual and worth stating plainly. A prompt of 199,000 tokens and a prompt of 201,000 tokens differ by one percent in size and by one hundred percent in price per token. On a pipeline that stuffs a context window until it fits, that boundary is where the money goes, and trimming a prompt back under it halves what the call costs.
Two thousand extra tokens can double the price of a call. On Grok, prompt length is a billing decision, not just an engineering one.
The output ratio, and why it flatters Grok
Output costs three times input on grok-4.6. That is the tightest ratio among the major providers: Anthropic charges five times more for output across the Claude line, and OpenAI runs at six times on its current models.
It matters because the ratio, not the input price, is what decides your bill on generative work. Two providers with the same input rate can be far apart once the model starts writing. Grok is priced for work that reads a lot and writes a lot, which is exactly the profile the 200k threshold then penalises, so the two effects pull against each other and the only way to settle it is on your own traffic.
The rest of the xAI line
Grok 4.6 is not the cheapest model xAI sells, and on prompts under 200k it is not even cheaper than the version before it. To understand how its pricing compares to newer releases, you can check what grok-4.7 costs per million.
| Model | Input / cached / output, under 200k |
|---|---|
| grok-4.6 | $2.00 / $0.50 / $6.00 |
| grok-4.5 | $2.00 / $0.30 / $6.00 |
| grok-4.3 | $1.25 / $0.20 / $2.50 |
| grok-build-0.1 | $1.00 / $0.20 / $2.00 |
Grok 4.6 and grok-4.5 carry identical input and output rates. The only line that separates them is cached input, where 4.5 is cheaper at $0.30 against $0.50. If your workload is dominated by cache reads, that is a real and counterintuitive difference: the older model has the cheaper cache.
Below them, grok-4.3 is priced at $1.25 and $2.50, which is less than half of 4.6 on output, and grok-build-0.1 sits lower still. Both keep the same doubling past 200k, at $2.50 and $5.00 for 4.3.
What a month actually costs
An agent workload of twenty million input and four million output tokens with no caching, priced three ways to show what the threshold does.
| Share of traffic over 200k tokens | A month on grok-4.6 |
|---|---|
| None, every prompt short | $64.00 |
| A quarter of it | $80.00 |
| All of it | $128.00 |
Same tokens, same model, twice the bill. No other provider in our comparison makes prompt length swing a month this hard, which is why a flat per-million figure for Grok is close to meaningless without knowing the traffic behind it. Our comparison of what every major model charges uses the short-prompt rate, and says so.
How you buy it changes the price
Everything above is the rate at the xAI API. Routing through an aggregator adds a margin on top of the provider grid, which we measured on OpenRouter, and on a model whose price already doubles on long prompts, a few percent of markup lands on the higher of the two grids rather than the lower one. If your traffic is long, check what the router charges on the long grid specifically, not on the headline rate.
For the wider view, our full Grok API pricing page covers the rest of the xAI catalogue, and the LLM cost calculator prices the same volume across every provider we track.
Frequently asked questions
How much does Grok 4.6 cost per million tokens?
$2.00 input and $6.00 output on prompts under 200,000 tokens, with cached input at $0.50. Past 200,000 tokens the rates are $4.00, $12.00 and $1.00. Read from the xAI model documentation on 16 August 2026.
What happens at 200,000 tokens exactly?
Every billing line doubles, cached input included. The threshold applies per prompt, so a single long call is billed on the high grid while the rest of your traffic stays on the low one.
Is grok-4.6 more expensive than grok-4.5?
Only on cached input, where 4.6 charges $0.50 against $0.30. Fresh input and output are identical at $2.00 and $6.00.
What is the cheapest xAI model?
Of the models we read on that page, grok-build-0.1 at $1.00 input and $2.00 output, followed by grok-4.3 at $1.25 and $2.50. Both double past 200,000 tokens like the rest of the line.
How does Grok compare with Claude and GPT on price?
On the short-prompt grid, grok-4.6 sits close to Claude Sonnet 5 on input and above it on output, and its three-times output ratio is tighter than either Anthropic’s five or OpenAI’s six. We put them side by side, read on the same day, in our cross-provider comparison.
What we verified and what we did not
Verified: input, cached input and output rates for grok-4.6, grok-4.5, grok-4.3 and grok-build-0.1, on both the under-200k and the 200k-and-over grids, read from the xAI model documentation on 16 August 2026.
Not verified, and therefore not asserted: whether the threshold counts input tokens alone or the full context including output, which the page does not spell out; any batch or volume discount, which xAI does not publish on that page; and any benchmark or quality claim, since this page prices models rather than ranking them.
Prices in this category move. Everything here carries the date it was read, and the grid is worth checking at the source before you commit a budget to it.