Grok 4.7 became the newest xAI model at the end of September 2026. According to the official docs.x.ai documentation, it is available at the same time in the xAI API, in the Grok Build terminal agent and in the Cursor editor. The GitHub blog separately confirmed its arrival in GitHub Copilot on 21 September 2026, on the Pro, Pro+, Max, Business and Enterprise plans. Grok 5, despite the regular rumors, had not shipped by then.
What is new in Grok 4.7
The official model documentation presents Grok 4.7 as xAI's most capable model for code, agentic tasks and knowledge work. The main characteristics:
- a context window of 500,000 tokens;
- input: text and image, output: text;
- four reasoning levels:
low,medium,high(the default) andxhigh; - training data cutoff: May 2026.
API pricing: $2 per million input tokens and $6 per million output tokens when the prompt is no longer than 200 thousand tokens. Above that threshold the rates rise to $4 and $12. The full API price list for the line is covered in Grok API: models, prices and a key.
How 4.7 differs from 4.6 and 4.5
On price and context the models match: Grok 4.5, 4.6 and 4.7 share the same 500 thousand token window and the same $2/$6 rate per million tokens. The revisions differ in positioning and targeted refinements, not in the architectural parameters xAI itself publishes.
xAI announced Grok 4.5 as a model trained alongside Cursor ("trained alongside Cursor"), and on release day it opened on every Cursor plan. The company called Grok 4.6 its flagship for long agentic tasks and visual scenarios, and for the first week after launch it doubled the included usage in Grok Build and Cursor. Grok 4.7 continues that line with a focus on code and "knowledge work" and ships in four places at once: the API, Grok Build, Cursor and GitHub Copilot.
In practice that means the following. If you already work in Cursor or Grok Build on Grok 4.6, moving to 4.7 needs no change of infrastructure: the new model shows up in the same selector.
What Heavy is
Grok Heavy is not a separate model but a high-compute mode. Normally one agent handles a request to Grok (or up to four, according to third-party market reviews). In Heavy mode up to 16 agents work on a single request in parallel and then cross-check the results. It needs a SuperGrok Heavy subscription, the most expensive plan in the line. For the plans and what really works from Russia, read Grok in Russia: the SuperGrok subscription.
Has Grok 5 been released
No. As of 25 September 2026, a search of xAI's official news section and of the model documentation turns up no Grok 5 announcement. The press reports that the release queue shifted: extra 4.x revisions, including the already released 4.7, were added to the plan ahead of the fifth version. That fits xAI's usual habit of shipping frequent intermediate versions instead of rare big ones: over 2026 the company took the line from 4.3 to 4.7 without announcing a fifth generation.
If you come across a specific Grok 5 release date, treat it as a forecast, not a fact. When this piece was prepared, as earlier in 2026, no official announcement existed.
A word on the company itself. In July 2026 xAI rebranded after merging with SpaceX, and the parent structure came to be called SpaceXAI (several independent outlets wrote about it, including Yahoo Finance and Techzine). The Grok brand and the x.ai and docs.x.ai domains stayed the same, so users of the API and the products barely noticed the shuffle.
Grok 4.7 or Claude Opus 5 in Cursor and Claude Code
When you pick a model for everyday work in Cursor, you are weighing Grok 4.7 at $2/$6 per million tokens against Claude models through a third-party API such as codexkey. xAI and Anthropic do not publish a direct benchmark comparison on one methodology, so instead of invented numbers look at the practical difference. Grok 4.7 has the narrower context window (500K against 1M on some Claude models), but for certain tasks its reasoning ceiling in xhigh mode is higher.
For agentic coding in the terminal xAI has Grok Build, a direct competitor to Claude Code. If you are used to Claude Code and only want to swap the model provider under the hood without learning a new tool, with codexkey you can do it without rebuilding your workflow:
export ANTHROPIC_BASE_URL="https://api.codexkey.online"
export ANTHROPIC_AUTH_TOKEN="sk_cdx_YOUR_KEY"
export ANTHROPIC_DEFAULT_OPUS_MODEL="claude-opus-5-5[1m]"
export ANTHROPIC_DEFAULT_SONNET_MODEL="claude-sonnet-5[1m]"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-sonnet-5"
The ANTHROPIC_DEFAULT_* lines are mandatory: without them Claude Code asks for its own default models and answers "There's an issue with the selected model". After that Claude Code works as usual, except that requests go through codexkey, payment is in rubles, and there is no five-hour subscription window: you pay for the tokens you actually use. Full setup for each OS is in the Claude Code guide. If you are comparing several agentic tools rather than only Cursor and Claude Code, there is an overview in Codex, Cursor, Claude Code: a comparison.
To be honest: if your process is already tied to the xAI ecosystem, or you specifically need Grok Imagine and the Grok Voice assistant, moving to Claude through codexkey will not solve those jobs. But for writing and refactoring code, pull request review and agentic terminal scenarios, the gap between the models is in most cases smaller than the gap in how easy it is to pay and get access from Russia.
FAQ
When was Grok 4.7 released?
According to the GitHub blog, the GitHub Copilot integration happened on 21 September 2026, together with availability in the xAI API, Grok Build and Cursor. We could not find the exact date of the first release with a publication date on xAI's official site in open sources.
Has Grok 5 been released?
No. As of 25 September 2026 Grok 5 had not shipped. No release date announcement appears on any xAI page we checked.
How does Grok 4.7 differ from Grok 4.6?
By the officially published parameters, context and price, the models are the same. The emphasis differs: xAI calls 4.7 its flagship for code and knowledge work, while 4.6 was aimed at long agentic tasks and visual scenarios.
What is Heavy mode in Grok?
It is not a separate model but a high-compute mode in which up to 16 agents process one request in parallel and cross-check the results. It is available with a SuperGrok Heavy subscription.
Is Grok 4.7 available in Cursor?
Yes. According to the official model documentation, Grok 4.7 has been available in Cursor since launch, alongside Grok Build and the xAI API.
Is it true that xAI was renamed?
After the merger with SpaceX in July 2026 the parent company came to be called SpaceXAI, as independent outlets reported. The Grok brand and the x.ai domain for documentation and the API stayed unchanged.