Grok 4.7: specs, pricing, the 200k price step, and where your prompts go

The Morse teamPublished October 25, 20268 min read

Grok 4.7 is SpaceXAI's closed-weight frontier model for coding, agent work, and knowledge work, released on September 21, 2026. As of October 25, 2026, it has a 500,000-token context window, takes text and images and answers in text, offers four reasoning settings, and costs $2 per million input tokens and $6 per million output tokens on the API until a prompt reaches 200,000 tokens, when the price for the whole request doubles. SpaceXAI LLC runs Grok and the xAI API.

We read SpaceXAI's Grok 4.7 model page, its Grok 4.7 overview, the API release notes, the launch post, the API page, the subscription pricing, the security FAQ, the consumer FAQ, and the privacy policy on October 25, 2026. Benchmark scores are SpaceXAI's unless we name another tester. SpaceXAI's developer docs are more complete than most labs'. One page gives the model's rate limits, regions, knowledge cutoff, and both price tiers, and the security FAQ says plainly that the API doesn't train on your prompts.

Grok 4.7 specs at a glance

The launch post says Grok 4.7 "uses a new, larger base model compared to Grok 4.6," trained with a longer reinforcement learning run "weighted toward problems that take many hours to complete" (SpaceXAI). It gives no parameter count. On the Responses API, every Grok 4.7 reply carries its reasoning back to you in encrypted form, which you pass back unchanged in the next turn so the model keeps its train of thought (SpaceXAI).

How much does Grok 4.7 cost?

As of October 25, 2026, the API charges by the token in two tiers (SpaceXAI):

The step applies to the whole request. In SpaceXAI's words, "requests whose prompt reaches 200k tokens are billed at the higher rate for all tokens in the request," output included. Here is what one request with 5,000 tokens of output costs at each size, by our arithmetic from those rates, with no cached input or tool calls:

Going from 199,000 to 200,000 tokens of prompt doubles the bill for 1,000 more tokens. If you send long documents or long agent histories, trimming them under 200,000 tokens halves the price of every request. SpaceXAI's docs list two discounts and one premium: the Batch API is 20% cheaper (SpaceXAI), cached input is a quarter of the input price, and the US regional endpoint at us.api.x.ai, which keeps inference in the United States, costs 10% more (SpaceXAI).

Long replies add to the cost. Artificial Analysis, which tests models independently, found Grok 4.7 at xhigh effort "very verbose." It wrote 240 million output tokens to run the firm's Intelligence Index, against a median of 87 million, and averaged $3.74 a task (Artificial Analysis). Lower effort settings write less.

Is Grok 4.7 free?

Only in places. SpaceXAI's API page answers that API usage "is billed per token rather than free," though every Console account gets a Playground for testing (SpaceXAI). The Grok Build coding agent, where Grok 4.7 is the default model, is "available to try for free" (SpaceXAI), and Cursor offers Grok 4.7 on all its plans (SpaceXAI).

On Grok.com, Grok 4.7 sits behind the mode menu. Grok.com's plans page, viewed signed out on October 25, 2026, lists Fast as available, and labels Expert ("Thinks hard") and Heavy ("Team of Experts") as running Grok 4.7, both marked as needing an upgrade (Grok). The launch post itself names Cursor, Grok Build, and the API, and doesn't mention the Grok app. As of October 25, 2026, The X Subscription, which now bundles Grok, Cursor, and X, costs $0 (Free, "Limited AI usage"), $8 (Plus), $30 (Premium, which lists "Latest Grok models"), $100 (Super), or $200 (Ultra) a month, with usage drawn from a pool that refreshes weekly (SpaceXAI).

Grok 4.7 vs GPT-5.6 Sol and Fable 5.1

Grok 4.7 is cheaper per token than both and wins some of SpaceXAI's tests. The launch post ran Grok 4.7 at xhigh effort and the other two at max, and lists their prices as $4 and $20 per million tokens for GPT-5.6 Sol and $10 and $50 for Fable 5.1 (SpaceXAI):

SpaceXAI chose these tests and ran them, so read them as its claims. Artificial Analysis scores Grok 4.7 at 46 on its Intelligence Index, 31st of the 215 models it ranks, at 92.7 output tokens a second, with an average wait of 40.25 seconds for the first token (Artificial Analysis).

Where do your Grok 4.7 prompts go?

To SpaceXAI either way, under different rules. The privacy policy, last updated October 11, 2026, covers Grok.com and the Grok apps, and says it "does not apply to data that we process on behalf of customers of our business offerings," naming the API as one (SpaceXAI). The same model gets two sets of data rules:

On Grok.com and the apps

Grok.com and the apps run on the consumer rules in the table, so Grok 4.7 gets the same training switch, Private Chat, and 30-day deletion as any Grok chat, which we cover in is Grok private. Grok inside X follows X's privacy policy instead of SpaceXAI's, the privacy policy says.

On the API

The API trains on nothing by default and keeps a 30-day copy unless a team admin turns on Zero Data Retention in the Console (SpaceXAI). If the option doesn't appear, the FAQ says self-serve zero retention "may not be enabled for your environment yet," or the team's negotiated enterprise agreement may govern retention instead, and points you to sales@x.ai. Turning it on also switches off the Batch API and its 20% discount; we compare xAI's terms with other providers' in what zero data retention covers. Using the API at all takes a Console account, a key, billing, and code or an app that calls it.

How to use Grok 4.7 without a SpaceXAI account keeping your prompts

Every row in the table above depends on an account with SpaceXAI: a consumer account that trains by default and keeps chats until you delete them, or an API team you set up, fund, and configure for zero retention yourself. On Morse, Grok 4.7 is one of the Pro text models, unlimited on Pro at $19 a month, and you pick it from a menu with no SpaceXAI account, no API key, and no Grok.com login.

• Private, to run Grok 4.7 under zero data retention. Morse runs xAI's Grok in Private mode, the mode a new chat starts in. Our proxy removes your account, email, and IP address and sends the message under Morse's own account, and Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't; otherwise it relies on the provider's commitment. There's no "Improve the model" switch to find and no Console setting to enable.
• Anonymous, for GPT and Claude, to check SpaceXAI's benchmark table. In the table above, Grok 4.7 beats GPT-5.6 Sol on three of the five tests and Fable 5.1 on two. Run Grok 4.7 in a Private chat, then put the same prompt to GPT or Claude in Anonymous mode. The provider sees the message without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy.
• E2EE, for prompts no company should read. Grok is closed-weight and doesn't run in E2EE, so for those chats pick an open-weight model like GLM, Kimi, or DeepSeek. Your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay ciphertext they can't read. The guarantee rests on the enclave hardware and attestation working as designed. E2EE chats come with Pro too; see how end-to-end encrypted AI chat works.

In every mode, your chat history is stored in your browser, encrypted, not on Morse's servers, and it locks when you sign out. Morse doesn't train AI on your content, sell your data, or show ads (privacy policy), and because it doesn't store chats, it usually has very little to give authorities. Clearing your browser or deleting your account deletes your chats, and Morse can't recover them, so keep a copy of anything you need. If you're leaving Grok.com for other reasons, our Grok alternatives page compares the options.

Start free, then choose Pro at $19 a month for unlimited Grok 4.7 in Private mode and end-to-end encrypted chats with open-weight models.