Grok 4.6 and Grok 4.7: The New Flagship Models — Grok in 2026: From Grok 4.5 to 4.7, Grok Bot, and The X Subscription

Grok 4.5, introduced in previous lessons as the summer flagship model, has been outpaced twice in two months. Grok 4.6 arrived on August 12, 2026, followed by Grok 4.7 on September 21. Official pages now bear the SpaceXSI name, while developer documentation continues to refer to the “xAI API”: these are the same models and the same API.

The Two Models at a Glance

Feature	Grok 4.6	Grok 4.7
ID	`grok-4.6`	`grok-4.7`
Announcement	August 12, 2026	September 21, 2026
Context	500,000 tokens	500,000 tokens
Input / Output	Text and image / text, no output limit	Text and image / text, no output limit
Price under 200,000 prompt tokens	$2 / $0.50 (cache) / $6	$2 / $0.50 (cache) / $6
Price over 200,000 tokens	$4 / $1 / $12	$4 / $1 / $12
Reasoning Effort	`low`, `medium`, `high` (default), `xhigh`	`low`, `medium`, `high` (default), `xhigh`

The models page is clear: for anything outside audio, image, or video—including code—use Grok 4.7, billed as the editor’s most capable model. Its knowledge cuts off in May 2026. Recall the tiered pricing rule seen with Grok 4.5: once a prompt hits 200,000 tokens, the entire request moves to the higher rate.

The xAI documentation Models page: Grok 4.7 recommended for code and conversation, with Voice and Imagine APIs — The official models page from the xAI documentation (docs.x.ai)—captured on November 2, 2026.

What Changes in Your Code

Switching from grok-4.5 to grok-4.7 usually comes down to the ID, but three details are worth noting. First, on the Responses API, grok-4.7 always returns encrypted reasoning (reasoning.encrypted_content), even if you didn’t request it in include: account for this if you store or log responses. Next, logprobs and top_logprobs are not supported from grok-4.20 onward and are silently ignored. Finally, Grok 4.7 is also served on a US regional endpoint, useful if your data residency constraints require it.

from openai import OpenAI
import os

client = OpenAI(api_key=os.environ["XAI_API_KEY"], base_url="https://api.x.ai/v1")

response = client.responses.create(
    model="grok-4.7",
    input="List the five main risks of a production database migration.",
    reasoning={"effort": "high"},
)
print(response.output_text)

A variant exists, Grok 4.7 Fast: the same model, faster, billed at twice the standard rate (one and a half times for long context). It is not offered on the public API, only in Cursor and Grok Build, where it is billed within the subscription.

Grok at Major Providers

Grok 4.6 deployed within days across several providers: in GitHub Copilot on August 14, generally available on Amazon Bedrock on August 19, on Google’s Gemini Enterprise Agent Platform on August 21, and on Microsoft Foundry on August 26. For an organization that already has a contract with one of these providers, this is often the simplest path to test Grok without opening a new account. Check each platform for available model versions and their update dates, which follow the provider’s schedule.

Official Sources: Models · Release notes · Introducing Grok 4.7

Pricing captured on November 2, 2026 — check the current rate card on the xAI models page.

Key Takeaways

Grok 4.7 (grok-4.7, September 21, 2026) is the recommended model for text and code; Grok 4.6 preceded it on August 12
500,000 context tokens, $2 / $6 per million tokens under 200,000 prompt tokens, $4 / $12 over
Reasoning effort low to xhigh, high by default; encrypted reasoning always returned on the Responses API
Grok 4.6 is available in GitHub Copilot, on Amazon Bedrock, Gemini Enterprise Agent Platform, and Microsoft Foundry