How much does each Grok API model cost?
Grok 4.7 costs $2 per million input tokens and $6 per million output tokens, the same as grok-4.6, and it tops xAI's list of eight text models. We read every row in this table from xAI's API pricing page on 8 November 2026. All of the prices are in US dollars per million tokens, and we worked out the Batch column ourselves from the 20% off that xAI states:
| Model, read 8 November 2026 | Context | Input | Cached input | Output | From 200K prompt tokens: in, cached, out | Batch: in, out |
|---|---|---|---|---|---|---|
| grok-4.7 | 500K | $2.00 | $0.50 | $6.00 | $4.00, $1.00, $12.00 | $1.60, $4.80 |
| grok-4.6 | 500K | $2.00 | $0.50 | $6.00 | $4.00, $1.00, $12.00 | $1.60, $4.80 |
| grok-4.5 | 500K | $2.00 | $0.30 | $6.00 | $4.00, $0.60, $12.00 | $1.60, $4.80 |
| grok-4.3 | 1M | $1.25 | $0.20 | $2.50 | $2.50, $0.40, $5.00 | $1.00, $2.00 |
| grok-4.20-0309-reasoning | 1M | $1.25 | $0.20 | $2.50 | $2.50, $0.40, $5.00 | $1.00, $2.00 |
| grok-4.20-0309-non-reasoning | 1M | $1.25 | $0.20 | $2.50 | $2.50, $0.40, $5.00 | $1.00, $2.00 |
| grok-4.20-multi-agent-0309 | 1M | $1.25 | $0.20 | $2.50 | $2.50, $0.40, $5.00 | $1.00, $2.00 |
| grok-build-0.1 | 256K | $1.00 | $0.20 | $2.00 | $2.00, $0.40, $4.00 | No discount |
For almost anything you'd build, xAI tells you to pick the newest model. Its models page says "For everything else, including code, use Grok 4.7." It costs the same as grok-4.6, so there's no reason to stay on the older one. The only case for grok-4.5 is that its cached input is cheaper, at $0.30 per million. If you want cheap tokens, the grok-4.3 tier cuts the output price by more than half. And grok-build-0.1 is the cheapest of all.
Reasoning isn't free. xAI's caching docs list reasoning tokens at the "Full completion token price." On grok-4.7 the default effort is high, so even a short answer can come with a long think that you pay for.
What happens to the price past 200K tokens?
The price doubles for the whole request once the prompt hits 200,000 tokens. xAI says requests whose prompt reaches the threshold "are billed at the higher rate for all tokens in the request." So it's not just the tokens above 200K. Here's what the input of one prompt costs you on either side of the line, as of 8 November 2026:
The output doubles too. On grok-4.7, a 2,000-token answer costs $0.012 below the line and $0.024 above it. If your prompts sit near 200K, trim them before you send them, or use the context compaction that xAI suggests for long agent loops. The 1M context on grok-4.3 doesn't dodge this. It just lets you go further past the line. You'll still pay double.
What do Grok API tools cost?
Grok's server-side tools cost $5 per 1,000 calls for web search and code execution, and on top of that you pay for all of the tokens they pull in. They're all on xAI's pricing page, as of 8 November 2026:
| Tool | What it does | Price |
|---|---|---|
| Web Search | Searches and reads web pages | $5 per 1,000 calls |
| X Search | Searches X posts, profiles and threads | $5 per 1,000 posts, $10 per 1,000 profiles |
| Code Execution | Runs Python in a sandbox | $5 per 1,000 calls |
| File Attachments | Searches files attached to a message | $5 per 1,000 calls |
| Collections Search | Queries your uploaded documents | $2.50 per 1,000 calls |
| Image Generation | Makes and edits images | Imagine rates |
| Image and X video understanding | Reads media found by search | Tokens only |
| Remote MCP tools | Calls your own MCP servers | Tokens only |
Each search is half a cent. That's as much as 2,500 input tokens on grok-4.7. Anthropic's pricing page charges $10 per 1,000 searches, and OpenAI's API price list charges $10 per 1,000 calls, so on Grok it's half the price.
Watch X Search. It's billed "per item fetched rather than per call," and every post that comes back counts, with parent and quoted posts included. A search that brings back 20 posts costs $0.10, which is more than 18 typical support-bot replies on grok-4.7.
What does a typical month on the Grok API cost?
A support bot that answers 50,000 questions a month costs about $270 on grok-4.7 and $132.50 on grok-4.3. We built each request out of a 2,000-token system prompt that hits the cache, plus 1,000 new input tokens and a 400-token answer. Prices are xAI's list rates read on 8 November 2026:
| Setup | Per request | 50,000 requests a month |
|---|---|---|
| grok-4.7, standard | $0.0054 | $270 |
| grok-4.7, no cache hits | $0.0084 | $420 |
| grok-4.7, Batch API | $0.00432 | $216 |
| grok-4.7, Priority | $0.0108 | $540 |
| grok-4.7, US endpoint | $0.00594 | $297 |
| grok-4.3, standard | $0.00265 | $132.50 |
| grok-build-0.1, standard | $0.0022 | $110 |
Here's the grok-4.7 row in full. The 2,000 cached tokens come to $0.001 at $0.50 per million. The 1,000 new tokens add $0.002, and the 400 tokens of the answer add $0.0024. That's $0.0054 for each request, or $270 for all 50,000.
In this case the cache is worth $150 a month to you. xAI says to set a prompt_cache_key, because without one "you often pay full input price on a cache-cold server." Batch takes 20% off, but most jobs finish "within 24 hours," so it suits evals and backfills, not a live bot. Priority doubles every token type, and the US endpoint adds 10%.
Two things push it up. If one request in five runs a web search, that's 10,000 calls and $50 more. And if grok-4.7 thinks for 1,500 tokens before each answer, those bill at $6 per million: $0.009 a request, or $450 more a month. Set the reasoning effort low on easy questions.
Is the Grok API cheaper than OpenAI and Claude?
Yes for most chat and support work, because grok-4.7's output costs $6 per million against $10 for GPT-6.1 Sol and Claude Sonnet 5.5. We read all three price lists on 8 November 2026, from xAI, OpenAI and Anthropic:
| Model, read 8 November 2026 | Input | Cached input | Output | Web search per 1,000 |
|---|---|---|---|---|
| grok-4.7 | $2.00 | $0.50 | $6.00 | $5 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 | $10 |
| Claude Sonnet 5.5 | $2.00 | $0.10 | $10.00 | $10 |
| grok-4.3 | $1.25 | $0.20 | $2.50 | $5 |
| GPT-6.1 Terra | $1.00 | $0.05 | $5.00 | $10 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 | $10 |
| Claude Haiku 5.5, prompts to 100K | $0.10 | $0.01 | $0.50 | $10 |
On the support bot above, GPT-6.1 Sol and Sonnet 5.5 both cost $0.0062 a request, or $310 a month, before the cache-write fees that OpenAI and Anthropic list and xAI doesn't. That makes grok-4.7 $40 a month cheaper.
Grok loses when the cache does most of the work. Its cached input costs five times as much as theirs. Take an agent turn with 50,000 cached tokens, 1,000 new ones and a 500-token answer. That's $0.030 on grok-4.7 and $0.012 on Sonnet 5.5, so Grok costs 2.5 times as much. It also has no budget tier: the cheapest Grok model is ten times the input price of GPT-6 Luna or Claude Haiku 5.5. ChatGPT API vs Claude API pricing compares those two labs in depth.
What do Grok Imagine and Grok Voice cost on the API?
Grok Imagine costs $0.02 or $0.04 for an image and from $0.02 to $0.08 for each second of a video, as of 8 November 2026. Grok Voice costs $0.08 a minute when it's speech to speech. Here's what a few common jobs cost at those rates:
| Job | Model | Rate | Cost |
|---|---|---|---|
| 100 images | grok-imagine-image-2.0 | $0.04 an image | $4.00 |
| 100 images | grok-imagine-image | $0.02 an image | $2.00 |
| One 10-second clip | grok-imagine-video-1.5 | $0.08 a second | $0.80 |
| One 10-second clip | grok-imagine-video | $0.05 a second | $0.50 |
| One 10-second clip | grok-imagine-video-1.5-lite | $0.02 a second | $0.20 |
| One hour of voice chat | grok-voice-3.0 | $0.08 a minute | $4.80 |
| One hour of transcription | Speech to Text, REST | $0.10 an hour | $0.10 |
| 1 million characters read aloud | Text to Speech | $15 per 1M characters | $15.00 |
If you send image and video jobs through Batch, you still pay the full price for them. xAI says "Image and video generation are supported in the Batch API but are billed at standard rates." On 2 November, xAI retired grok-imagine-image-quality and now serves it as grok-imagine-image-2.0 on low quality, "at a lower per-image price."
What other fees and limits come with the Grok API?
The Grok API bills you from prepaid credit by default, and there are a few smaller fees on top of the tokens. Here's each one from xAI's docs, read 8 November 2026:
| Fee or rule | What xAI charges |
|---|---|
| Batch API | 20% off every token type on grok-4.7 down to the grok-4.20 models; none on grok-build-0.1 |
| Priority Processing | 2x every token type, billed only when the reply says "priority" |
| US regional endpoint | 1.1x, grok-4.7 and grok-4.6 only |
| File storage | $0.025 per GiB a day |
| Collection storage | $0.10 per GiB a day |
| File and collection downloads | $0.20 per GiB |
| Request blocked before generation | $0.05 usage guideline fee |
| Grok 4.7 Fast | 2x standard below 200K tokens and 1.5x above, in Cursor and Grok Build only, not on the API |
Your rate limits grow with what you've spent. On xAI's rate limits page, a team moves up to Tier 1 once it has spent $50 since 1 January 2026. It reaches Tier 2 at $250, Tier 3 at $1,000 and Tier 4 at $5,000. Monthly invoicing "is disabled by default," per the billing docs, and you have to ask sales for it.
What does the Grok API cost next to the rest of your AI stack?
The Grok API is a separate bill from any chat plan, and xAI doesn't put a dollar figure on the API use its plans include. On grok.com, "Grok API" is one of the benefits listed for Plus at $8 a month, with no stated amount. Anthropic's Max plan includes "$100 a month in API credits," and OpenAI's Pro plans include "Up to $500 in monthly API credits." Grok pricing covers the consumer plans.
So if you build on Grok and also chat with ChatGPT and Claude, you pay three ways: the API bill, $20 for ChatGPT Plus and $20 for Claude Pro. Whizi, which publishes this page, only touches the last two. Whizi Pro costs $29.99 a month and puts Claude, GPT, Gemini and Grok 4.7 in one chat. It has no API, no keys and no coding agent, so all of your app's traffic stays with xAI. That part doesn't change.
- Default to grok-4.7, which costs the same as grok-4.6
- Keep prompts under 200,000 tokens, or the whole request bills at double
- Set a prompt_cache_key so your system prompt bills at the cached rate
- Send evals and backfills through the Batch API for 20% off
- Cap X Search results, since every returned post costs $0.005
- Price cache-heavy agents on Claude Sonnet 5.5 too, where cached input costs a fifth as much
Frequently asked questions
Is the Grok API free?
No. xAI's quickstart tells new accounts to sign up and "load it with credits," and its pricing docs list no free tier as of November 2026. Even the cheapest text model, grok-build-0.1, costs you $1 for each million input tokens and $2 for each million output tokens. LLM API pricing statistics shows where that sits across labs.
How much does Grok 4.7 cost per million tokens?
Grok 4.7 costs $2 per million input tokens, $0.50 per million cached input tokens and $6 per million output tokens, for prompts under 200,000 tokens, as of November 2026. From 200,000 tokens up, all three of those rates double, to $4, $1 and $12. ChatGPT API vs Claude API pricing gives the rival rates.
Is the Grok API cheaper than the OpenAI API?
Yes for output-heavy work: grok-4.7 charges $6 per million output tokens against $10 for GPT-6.1 Sol, at the same $2 input. It loses on cached input, at $0.50 against $0.10, and OpenAI's GPT-6 Luna costs a tenth of the input price of Grok's cheapest model. The cheapest way to use Claude and GPT covers the subscription side.
Does a Grok subscription include API access?
Partly. On grok.com, "Grok API" is listed among the benefits of Plus at $8 a month and above, but xAI doesn't say how much use that covers. Heavy API use is billed per token from prepaid credit in the xAI console. One subscription for all AI models explains what one chat plan can and can't replace.
