What It Costs to Run Grok 4.6
Grok 4.6 is billed per token, and the schedule reported at launch splits into two rate tiers at a 200,000 token prompt: a standard rate below the threshold and a higher one above it. This page does not quote the per token figures. xAI publishes the live ones in its own developer console, which is the only source that stays correct.
That is not a dodge. Grok 4.6 was reported to ship on August 12, 2026, and the public pricing detail available for it traces back to launch coverage rather than to a vendor page we could verify independently. Figures copied out of that coverage into a third party page age badly.
What does not move as fast is the shape of the bill, and the shape is what teams actually get wrong. That is what this page covers.
Model spend is also the one AI line item that other people will cover for you. AI Perks tracks $7.7M in credits across 194 companies, which is how most early teams fund a first year of inference.
The 200K Token Threshold, and Why It Matters
Launch coverage of Grok 4.6 describes two rate tiers split at a 200,000 token prompt. What that coverage does not settle is whether the higher tier applies to the whole request or only to the tokens above the line, and we found no plain language answer from xAI itself.
That ambiguity is the single most useful thing to know about Grok pricing, because the two readings produce very different bills. If the higher tier reprices the whole request, a prompt that crosses 200,000 tokens gets more expensive retroactively, on every token in the call. If it applies only above the line, crossing the threshold costs almost nothing at first. Budget for the worse reading until your own invoice tells you which one is real.
Three things worth designing around either way:
Trim hard near the line. If the threshold does reprice the whole request, cutting a few thousand tokens out of a prompt that sits just above 200,000 is among the cheapest optimisations in your stack. If it does not, you have still trimmed a prompt.
Watch conversations grow into it. Agent loops and long sessions accumulate context quietly. A session that crosses 200,000 tokens on turn 40 stays above the threshold for every turn after it, including the short ones.
Cache before you cross. xAI reportedly bills cached input well below fresh input, so a large stable prefix is the cheapest kind of big prompt. The cached rate is reported to be tiered at the same 200,000 token boundary.
Grok 4.6 is reported to carry a 500K context window, so there is a lot of room above the threshold. The window and the rate change sit in different places, which is exactly why the threshold is easy to blunder across.
Grok 4.6 vs Grok 4.5 vs Grok 4.3
xAI is reported to have shipped three flagship text models in roughly four months, and on the reported specifications they are not a clean upgrade ladder: the older Grok 4.3 is said to carry a larger context window than either model that followed it.
Every figure below comes from launch coverage rather than a vendor page we could verify.
| Model | Reported release | Reported context window | Published pricing |
|---|---|---|---|
| Grok 4.6 | August 12, 2026 | 500K tokens | Two reported tiers, split at a 200K token prompt |
| Grok 4.5 | July 8, 2026 | 500K tokens | Not isolated in any source we could verify |
| Grok 4.3 | Around April 2026 | 1M tokens | Not found in any source we could verify |
| Grok Imagine Image 2.0 | August 7, 2026 | Not applicable | Per image, by resolution and quality |
Grok 4.6 is positioned as an Opus class flagship tuned for long running agents, coding, research and visual work, expanding text, image and audio multimodality. Grok 4.5, reported about five weeks earlier, was also described as Opus class and as tuned on real developer sessions from Cursor, a training detail we could not confirm against anything published by xAI.
Grok 4.3 is the interesting one for document work. It is reported to offer a 1M token context window, native video input and document generation for PDFs, spreadsheets and slides. Coverage of the two newer models does not mention those capabilities, though that is an absence in the reporting rather than a confirmed removal. If your workload is one enormous input rather than a tight agent loop, the older model may still be the right call. Whichever you pick, the credit programs that offset the bill are the same ones listed at AI Perks.
Grok Imagine Image 2.0 Pricing
Grok Imagine Image 2.0 is billed per image rather than per token, which makes it far easier to forecast than the text models. Reported pricing runs from $0.04 to $0.08 per generated image, with about $0.01 added per input image on an edit.
These figures come from launch coverage rather than a vendor page we could verify, so confirm them in the console before budgeting against them.
| Output | Reported price per image |
|---|---|
| 1K resolution, low quality | $0.04 |
| 2K low quality, or 1K medium quality | $0.06 |
| 2K resolution, medium quality | $0.08 |
| Each input image on an edit | Adds about $0.01 |
At the cheapest reported tier, 10,000 generated images is roughly $400, which is a genuinely different budgeting conversation from token billing. The 2.0 release is reported to have added region level editing, multi image reference input of up to five images, Smart Resize across nine aspect ratios and better typography, and to have ranked second on public text to image and image editing arenas at launch.
What xAI Has Not Published
Three things founders ask about most are not documented in anything we could verify: a free API tier for Grok 4.6, a startup credit program, and any commitment to hold current prices.
No confirmed free API tier. Coverage refers to a rate limited free level in the consumer Grok product, but whether that extends to API access for Grok 4.6 is unclear in every source we checked. Assume you are paying from the first call.
No identified startup credit program. We could not identify a public xAI credit program for startups. One may exist unannounced, or may appear later, and AI Perks exists to catch it when it does.
No price stability commitment. We found no published undertaking from xAI to hold current rates. Anything you read about Grok pricing, including this page, is a snapshot.
Grok Imagine free access is reported gone. Third party pricing trackers report that Grok Imagine became paid only on March 19, 2026, alongside a new SuperGrok Lite tier at $10 per month covering short, low resolution generations, with standard SuperGrok at $30 per month or $300 per year. No top tier outlet confirmed that date directly, so treat the date and the subscription figures as reported rather than settled.
How to Cut a Grok API Bill
The largest saving available on any model API is not a prompt optimisation, it is not paying the bill at all. Credits cover model spend, compute spend, or both, depending on the program.
The credit layer comes first. The catalogue at getaiperks.com covers AI model providers and cloud infrastructure, which are the two categories where the bill for a Grok based product actually lands.
Credits at more than one provider are a hedge. Teams that build on a single model API carry all of the pricing uncertainty themselves. A second provider you already hold credit with is leverage as much as a discount.
Engineer around the 200K threshold. Summarise or truncate long agent histories before they cross it, and keep stable instructions in a cached prefix so the reported cache rate does the work.
Re-check the console monthly. Three flagship releases in roughly four months means published pricing on any third party page, this one included, ages fast.
Frequently Asked Questions
How much does the Grok 4.6 API cost?
Grok 4.6 is billed per token, and the schedule reported at launch has two rate tiers split at a 200,000 token prompt, with the higher tier above the threshold. This page does not quote the figures, because the reporting behind them is thin and xAI publishes current rates in its own developer console. Credits that offset model spend are tracked at getaiperks.com.
What is the Grok 4.6 context window?
Grok 4.6 is reported to have a 500K token context window, the same as Grok 4.5. The older Grok 4.3 is reported to carry 1M tokens, making it the larger window of the three despite being the earlier model. The reported rate threshold sits at 200,000 tokens regardless of window size.
Does xAI give free credits for the Grok API?
We could not identify a public xAI startup credit program, and we could not confirm a free API allowance for Grok 4.6 either way. Credits from other model and infrastructure providers still offset most of a Grok based product's real bill. See what is currently available at getaiperks.com.
How much does Grok Imagine Image 2.0 cost per image?
Reported pricing runs $0.04 per image at 1K low quality, $0.06 at 2K low or 1K medium, and $0.08 at 2K medium, with about $0.01 added per input image on an edit. These figures come from launch coverage rather than a verified vendor page, so confirm in the console.
Is Grok 4.3 or Grok 4.6 better for long documents?
For a single very large input, Grok 4.3 is reported to offer a 1M token window against 500K on Grok 4.6, so it fits more in one call. Grok 4.6 is positioned as the stronger choice for agent loops, coding and research work. Pricing for Grok 4.3 does not appear in any source we could verify, so price that side of the choice from your own console before committing.
Is Grok Imagine still free to use?
Reportedly not. Third party trackers say Grok Imagine moved to paid only on March 19, 2026, with a SuperGrok Lite tier at $10 per month covering short low resolution generations. xAI did not confirm that date in coverage we could verify, so check current access before planning around it.
Three flagship models in about four months, and no stable price to plan against. Let someone else pay for the tokens.