Grok API Pricing

Grok API Pricing: Every Model Rate, the 200k Rule, and What One Request Costs

On 5 November 2026, we reviewed xAI's developer pricing page and organized the listed charges into model rates, media rates, tool-call fees, and processing options. We also calculated the cost of one sample text exchange using 270 input tokens and 670 output tokens. The token counts came from our separate DeepSeek test, which lets readers compare the arithmetic on the same example. These calculations are estimates based on published rates, not charges from an actual Grok API invoice.

By Alex Pan · Checked: 5 November 2026 · xAI developer pricing documentation (docs.x.ai) · No API key created; the cost examples are computed from official rates, not from billed requests

At the rates listed on that date, grok-4.7 cost $2 per million input tokens and $6 per million output tokens; cached input was $0.50 per million. When a request reaches the 200,000-token prompt threshold, xAI says the higher rate applies to all tokens in that request. The listed grok-4.3 rates were lower ($1.25 input and $2.50 output per million) and its context window was 1 million tokens. The page also listed per-image, per-second video, voice, and tool-call charges. Below, we calculate a sample request and explain the extra charges that can change the total.

The text model rate table

The table below transcribes the text-model rates shown in xAI's developer pricing documentation on 5 November 2026. Rates are in US dollars per million tokens. Rows marked with the 200k threshold use the higher long-context rate when the request prompt reaches that threshold.

ModelContextInput / 1MCached input / 1MOutput / 1M
grok-4.7 (<200k prompt)500k$2.00$0.50$6.00
grok-4.7 (≥200k prompt)500k$4.00$1.00$12.00
grok-4.6 (<200k prompt)500k$2.00$0.50$6.00
grok-4.6 (≥200k prompt)500k$4.00$1.00$12.00
grok-4.5 (<200k prompt)500k$2.00$0.30$6.00
grok-4.5 (≥200k prompt)500k$4.00$0.60$12.00
grok-4.3 (<200k prompt)1M$1.25$0.20$2.50
grok-4.3 (≥200k prompt)1M$2.50$0.40$5.00
grok-4.20-0309 (reasoning / non-reasoning / multi-agent)1M$1.25$0.20$2.50
grok-4.20-0309 variants (≥200k prompt)1M$2.50$0.40$5.00
grok-build-0.1256k$1.00$0.20$2.00
xAI developer pricing page showing the Text API table: grok-4.7 at $2.00 input, $0.50 cached, $6.00 output per million tokens, with long-context rates at $4.00, $1.00 and $12.00
The Text API rate table on docs.x.ai, captured 5 November 2026. Left to right: context window, input, cached input, and output price per million tokens; the right-hand block is the long-context rate that applies once a prompt reaches 200k.

The models page listed May 2026 as grok-4.7's knowledge-cutoff date. It also said the model does not know about events or data beyond its training information unless server-side search tools are enabled. Those tools have separate charges, listed in the tool-pricing section below.

The 200k rule: one long prompt doubles every token in the request

The pricing page’s footnote explains how the long-context charge works:

“Models listed with two rows use long context pricing: requests whose prompt reaches the listed token threshold are billed at the higher rate for all tokens in the request.”

The important detail is that the higher rate applies to every token in a qualifying request, not only to tokens above 200,000. For example, a request with a 250,000-token prompt is not priced with the first 200,000 tokens at the lower rate and the remainder at the higher rate. Under the listed grok-4.7 rates, the higher tier is $4 per million input tokens and $12 per million output tokens. If your prompt is close to the threshold, reducing its size may lower the cost substantially. Check the current documentation before implementing a cost estimate, because thresholds and rates can change.

Images, video, and voice

Media generation is billed per output, not per token:

ModelPrice
grok-imagine-image-2.0$0.04 / image
grok-imagine-image$0.02 / image
grok-imagine-video-1.5-lite$0.020 / second
grok-imagine-video$0.050 / second
grok-imagine-video-1.5$0.080 / second
Speech to Speech (grok-voice-3.0)$0.08 / minute ($4.80 / hour) + $0.004 / text input
Speech to Text$0.10 / hour (REST), $0.20 / hour (streaming)
Text to Speech$15.00 / 1M characters

The documentation marks grok-voice-2.0 Speech to Speech as deprecated and lists grok-voice-3.0 at the same $0.08-per-minute rate. At $0.08 per second, a 15-second generation on the top video model in this table would cost $1.20 (15 × $0.08). This is an API calculation; it should not be treated as equivalent to a consumer subscription's video allowance.

Tool calls have their own price list

An agent request can generate both model-token charges and separate server-side tool charges. The token total can include reasoning tokens where the documentation identifies them as billable. The table below lists the tool-call rates shown on the pricing page.

ToolPriceHow it counts
Web Search (web_search)$5 / 1k callsIncludes browsing pages
X Search (x_search)$5 / 1k posts; $10 / 1k profilesPer item fetched, including parent and quoted posts
Code Execution$5 / 1k callsSandboxed Python
Collections Search (RAG)$2.50 / 1k callsQueries your uploaded document collections
File Attachments (attachment_search)$5 / 1k callsSearches files attached to messages
Image/video understandingToken-basedNo call fee; image tokens for processing
Remote MCP toolsToken-basedNo call fee; you pay tokens

For X Search, billing is based on the posts and profiles returned, not simply on the number of times the search button is pressed. The documentation says parent and quoted posts are included in the count, so retrieving a thread can create several billable items.

Batch, priority, and the US endpoint: three multipliers

One more rate exists but you cannot buy it on the public API: Grok 4.7 Fast, the same model on faster infrastructure, costs 2x standard rates (1.5x for long context) and is available only inside Cursor and Grok Build, billed through those plans. The docs state plainly that it “is not available on the public xAI API, and Grok Build’s free tier does not include it.”

What one request actually costs — next to DeepSeek

For a simple comparison, we used one sample exchange: 270 input tokens, 670 output tokens, no cache hit, and no tool calls. We multiplied those counts by each provider's published rates. The results are estimates for this one example, not typical costs for every prompt.

Service / modelInput rateOutput rateCost of this exchange
Grok grok-4.7$2.00 / 1M$6.00 / 1M≈ $0.0046
Grok grok-4.7, Batch$1.60 / 1M$4.80 / 1M≈ $0.0036
Grok grok-4.3$1.25 / 1M$2.50 / 1M≈ $0.0020
Grok grok-build-0.1$1.00 / 1M$2.00 / 1M≈ $0.0016
DeepSeek deepseek-flash, peak$0.30 / 1M$1.20 / 1M≈ $0.0009
DeepSeek deepseek-flash, off-peak$0.15 / 1M$0.60 / 1M≈ $0.0004
DeepSeek deepseek-v4-pro, peak$1.32 / 1M$3.96 / 1M≈ $0.0030

For this sample exchange, grok-4.7 comes to about $0.0046. That is roughly five times the listed peak-rate cost for DeepSeek flash in our comparison, and about 1.5 times the listed peak-rate cost for DeepSeek v4-pro. Batch pricing and cached input can reduce Grok's cost; DeepSeek's off-peak rate makes the comparison gap larger. With a cache hit on the 270 input tokens, the grok-4.7 example falls to about $0.0042.

These totals are calculated from the published rates and our measured 270/670 token counts. They are not actual invoices. Your cost will vary with prompt length, answer length, cache hits, tool calls, model choice, and any applicable processing multiplier.

Storage, downloads, and the violation fee

Three smaller line items from the same page that can add to a bill:

Frequently asked questions

Is there a free tier for the Grok API?
The developer pricing page we checked did not list a free token allowance or a free-credit offer. Grok Build has a free tier, but the documentation says it does not include Grok 4.7 Fast. Paid consumer plans from Plus upward include API access, but API usage is billed separately at model rates. We did not create an API key, so we cannot confirm whether a new account might receive credits through a separate promotion.
What is the Grok API base URL?
The global endpoint is https://api.x.ai/v1 and it is OpenAI-compatible, so the OpenAI SDK works by changing the base URL and key. The US regional endpoint is https://us.api.x.ai/v1 and bills at 1.1x standard rates with only grok-4.7 and grok-4.6 available.
How does prompt caching change the bill?
Cached input tokens bill at a fraction of the input rate — $0.50 instead of $2.00 per million on grok-4.7, $0.20 instead of $1.25 on grok-4.3. Caching discounts are applied before the batch, priority, and regional multipliers, per the docs.
Does a long prompt double the whole bill?
Yes. Once a request's prompt reaches the 200k-token threshold, the higher long-context rate applies to all tokens in that request, including output. A request just over the threshold costs about twice what the same request would cost just under it.
What does grok-4.7 know about current events?
Its knowledge cutoff is May 2026 per the docs, and there is no access to realtime events unless search tools are enabled. Web Search costs $5 per 1,000 calls and X Search is billed per post and per profile fetched.

Sources and what we could not test

ClaimWhere it came fromDate checked
Full text-model rate table, 200k long-context rule, media and voice ratesxAI developer pricing documentation5 Nov 2026
Tool prices (web/X search, code execution, collections), per-item X Search billingxAI developer docs, Tools Pricing section5 Nov 2026
Batch −20% model list, priority 2x with service_tier confirmation, US endpoint 1.1x, Grok 4.7 Fast restrictionsxAI developer docs, Batch / Priority / Regional sections5 Nov 2026
Storage $0.025/$0.10 per GiB/day, downloads $0.20/GiB, violation fee $0.05xAI developer docs, Files and Collections / Usage Guidelines5 Nov 2026
grok-4.7 May 2026 knowledge cutoff; no realtime events without search toolsxAI models page, Additional Information5 Nov 2026
DeepSeek comparison rates and the 270/670 token exchange shapeour own DeepSeek cost test (token counts measured Oct 2026; rates checked 2 Nov 2026)2 Nov 2026

We did not create an API key or send a paid request. As a result, we have no actual invoice, billing-dashboard record, rate-limit measurement, or latency test to report. The model list may also vary by region or account. Every cost in the worked example is calculated from the published rates; it is not an observed charge.

How this was researched: We read the xAI developer pricing documentation end to end on 5 November 2026 and transcribed every rate on it. The worked example reuses token counts from our own DeepSeek request test so the comparison is apples to apples. xAI did not pay us, and we do not sell anything on this page. See our full testing method.