Grok API price: cost of Grok 4.7, 4.6 and 4.5 with examples

If you are looking for the Grok API price, that is, how much requests to Grok cost through the API, the answer depends on the version, the ratio of input to output, and whether part of the context is read from cache. Zukko offers three versions: grok-4.7, grok-4.6 and grok-4.5. Below are a price table as of October 18, 2026, a formula and three calculations with arithmetic: a chat, a question about a long document with cache, and batch processing. Official xAI prices are shown alongside for comparison, so you can estimate the difference at your own volume.

Prices as of October 18, 2026

Prices in $ per 1M tokens:

The strength index is the Intelligence Index from Artificial Analysis. It helps to orient yourself but does not replace testing on your own tasks. Provider prices change, so this is an example for the stated date. The current list is GET https://api.zukko.pro/v1/models or the "Models" page in the dashboard.

All three versions have a 256K-token context (many catalog models have 1M), which matters if you plan to load large documents.

Formula

cost = fresh input × input price + cache read × read price + output × output price

All prices are per 1M tokens, so the total is divided by 1,000,000. About cache: the gateway counts cache reads when the provider reports them in the response. Cache writes for non-Claude models are not charged separately, so we do not use the "write" column in the calculations. Details on the rules: keys, limits and cache↗.

Example 1. Chat without cache

One request: 2,000 input tokens and 800 output tokens.

• grok-4.7: 2,000 × 0.175 + 800 × 0.513 = 350 + 410.4 = 760.4, i.e. $0.00076
• grok-4.6: 2,000 × 0.138 + 800 × 0.4 = 276 + 320 = 596, i.e. $0.000596
• grok-4.5: 2,000 × 0.113 + 800 × 0.338 = 226 + 270.4 = 496.4, i.e. $0.000496
• Official prices ($3 / $15): 2,000 × 3 + 800 × 15 = 6,000 + 12,000 = 18,000, i.e. $0.018

For $1 you get about 1 / 0.00076 ≈ 1,300 such requests on grok-4.7 and about 2,000 on grok-4.5 (1 / 0.000496 ≈ 2,016). This is price arithmetic, not a promise of quality.

Example 2. A question about a long document with cache

Suppose you ask many questions about one document: each request has 20,000 tokens of context, of which 18,000 are read from cache and 2,000 are fresh input, and the answer is 1,000 tokens. Assumption: the provider reports the cache read in the response; otherwise all input is charged at the input price.

• grok-4.7: 2,000 × 0.175 + 18,000 × 0.025 + 1,000 × 0.513 = 350 + 450 + 513 = 1,313, i.e. $0.001313
• grok-4.6: 2,000 × 0.138 + 18,000 × 0.013 + 1,000 × 0.4 = 276 + 234 + 400 = 910, i.e. $0.00091
• grok-4.5: 2,000 × 0.113 + 18,000 × 0.013 + 1,000 × 0.338 = 226 + 234 + 338 = 798, i.e. $0.000798

For comparison, the same request without cache on grok-4.7: 20,000 × 0.175 + 1,000 × 0.513 = 3,500 + 513 = 4,013, i.e. $0.004013. The cache lowers the bill because 18,000 tokens are charged at 0.025 instead of 0.175. At official prices the same request without cache: 20,000 × 3 + 1,000 × 15 = 60,000 + 15,000 = 75,000, i.e. $0.075. We do not give xAI's official cache price: it is not among our verified data, see the provider's website.

More detail on how the cache is calculated: prompt caching: read and write↗.

Example 3. A batch of 10,000 requests

Classification or short answers: 10,000 requests, 1,500 input tokens and 300 output each. In total 15,000,000 input (15M) and 3,000,000 output (3M).

• grok-4.7: 15 × 0.175 + 3 × 0.513 = 2.625 + 1.539 = $4.164
• grok-4.6: 15 × 0.138 + 3 × 0.4 = 2.07 + 1.2 = $3.27
• grok-4.5: 15 × 0.113 + 3 × 0.338 = 1.695 + 1.014 = $2.709
• Official prices: 15 × 3 + 3 × 15 = 45 + 45 = $90

The price difference between Grok versions in Zukko is a matter of dollars per 10,000 requests, while the difference in strength index (47, 44, 40) may matter more. If the lowest price for bulk tasks is what you need, look at other lines too: the cheapest models for bulk tasks↗.

How to estimate your own request in advance

You can estimate tokens before sending. For the Anthropic format there is POST /v1/messages/count_tokens. For Grok the simplest way is to send a small request, read the usage field in the response, and multiply the tokens by the table prices:

from openai import OpenAI

client = OpenAI(base_url="https://api.zukko.pro/v1", api_key="fetch_your_key")

PRICES = {"grok-4.7": (0.175, 0.513), "grok-4.6": (0.138, 0.4), "grok-4.5": (0.113, 0.338)}

def ask(model: str, text: str):
    resp = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": text}],
        max_tokens=300,
    )
    u = resp.usage
    p_in, p_out = PRICES[model]
    cost = (u.prompt_tokens * p_in + u.completion_tokens * p_out) / 1_000_000
    print(resp.choices[0].message.content)
    print(u.prompt_tokens, u.completion_tokens, f"${cost:.6f}")

ask("grok-4.6", "Explain the difference between a cache and model memory in three sentences")

This calculation ignores cache: if part of the input was read from cache, it is included in prompt_tokens in usage, and the actual price will be slightly lower than the calculated one. See the actual charge in the "Usage" section of the dashboard.

Grok API: how to buy access

The phrase "buy Grok API" means topping up a balance and getting a key. In Zukko: register on the site, top up from $1 (SBP with the page fee of 8.5% on top, cryptocurrency, MasterCard cards with a fee of about 13%, amounts $2–$594), and create a key in the API keys section↗. The fetch_... key works for all models in the catalog. On connecting Grok step by step: Grok 4.6 via API↗.

FAQ

How much does the Grok API cost per million tokens?

As of October 18, 2026 in Zukko, grok-4.7 costs $0.175 for input and $0.513 for output, grok-4.6 $0.138 and $0.4, grok-4.5 $0.113 and $0.338. Official prices for these versions: $3 for input and $15 for output.

Which Grok version is the better value?

By price the cheapest is grok-4.5, by strength index the highest is grok-4.7. Choose by a test on your own tasks: take 30–50 examples and compare the result with the price.

Does the cache work for Grok?

The gateway counts cache reads when the provider reports them in the response. Example 2 above shows how this affects the price. A separate cache write is not charged for Grok.

Does Grok accept images?

Yes, in the table image input is marked for all three versions. The context is 256K tokens.