Grok 4.7, Explained: Elon Musk's 2.1 Trillion-Parameter Claim
xAI released Grok 4.7 on September 21, 2026, after Elon Musk's target date slipped several times. This explainer separates what xAI confirmed from what is only a claim, and covers the pricing, where you can use it, and what a real workload would cost.
⚡ Quick facts
- Released: September 21, 2026, after Elon Musk's target date slipped several times
- Size: 2.1 trillion parameters according to xAI, up 40% from Grok 4.6's 1.5 trillion (not independently verified)
- API price: $2 per million input tokens and $6 per million output tokens, the same as Grok 4.6
- Where it runs: Grok app, xAI API, Cursor, Grok Build and GitHub Copilot (Pro, Pro+, Max, Business, Enterprise)
- Benchmarks: xAI-published scores trail Claude Fable 5.1, e.g. GDPval 1695 vs 1735
What launched
xAI, which now operates under the SpaceXAI name, released Grok 4.7 on September 21, 2026 (xAI announcement). It is a reasoning model aimed at agentic coding and long, multi-step workflows. xAI called it "a notable improvement over Grok 4.6 at the same price and speed", and says it reasons longer on hard problems and checks its own answers more often. xAI also says it added SpaceX data to training, including Starlink telemetry and engineering failure logs.
The road to launch
The release followed repeated delays. As Yahoo Tech tracked, Musk's public timeline changed at least five times after late July: "four weeks out", then "a few weeks", then "3 to 4 weeks", then "10 days" on September 1, then "needs a few more days to cook" on September 11. Days before launch, Musk himself set expectations lower, saying Grok 4.7 should roughly match Claude Opus 5.0 and that its multimodal performance still needed work.
What the 2.1-trillion-parameter claim tells you
xAI says Grok 4.7 has 2.1 trillion parameters, up 40% from Grok 4.6. Parameter count is a statement about size, not quality. It says nothing about how many parameters are active per token or how well the model was trained, and xAI has not published an architecture report. The useful numbers for a buyer are price, speed and results on your own tasks.
How it benchmarks
On the scores xAI published, Grok 4.7 is competitive but not the leader. It scored 1695 on GDPval against 1735 for Claude Fable 5.1, and 1657 on AA-Briefcase against 1678. On CursorBench it lands mid-chart, costing more per task than GPT-6 Astra and Claude Sonnet 5. These are vendor-reported numbers, so treat them as a starting point.
Where you can use it
Grok 4.7 launched without a waitlist in the Grok app, the xAI API, Cursor and Grok Build. On launch day GitHub added it to Copilot for Pro, Pro+, Max, Business and Enterprise plans, with a gradual rollout. It is selectable in VS Code, Visual Studio, JetBrains, Xcode, Eclipse, the Copilot CLI and the Copilot cloud agent, and billed at the provider's list price. Business and Enterprise admins can turn it off in Copilot's model policy.
Our take: a worked cost example
Say a coding agent reads 200,000 tokens of code and context per run and writes 20,000 tokens back.
- Input: 0.2 million tokens × $2 = $0.40
- Output: 0.02 million tokens × $6 = $0.12
- Per run: $0.52. At 1,000 runs a month, about $520.
Because the price did not change from Grok 4.6, switching costs nothing extra per token. The real question is whether it finishes your tasks in fewer runs. Pick 20 real tickets, run them on Grok 4.7 and on your current model, and compare pass rate and total spend, not leaderboard rank. If you use Copilot, check whether your organisation's model policy enabled it by default.
Sources
- xAI: Grok 4.7 announcement
- GitHub Changelog: Grok 4.7 in GitHub Copilot
- Yahoo Tech: xAI launches Grok 4.7
Frequently asked questions
When was Grok 4.7 released?
xAI released Grok 4.7 on September 21, 2026, after Elon Musk's public timeline slipped several times from late July onwards.
How many parameters does Grok 4.7 have?
xAI says 2.1 trillion, up 40% from Grok 4.6's 1.5 trillion. The figure has not been independently verified, and xAI has not published an architecture report.
How much does Grok 4.7 cost?
In the xAI API it costs $2 per million input tokens and $6 per million output tokens, the same as Grok 4.6.
Is Grok 4.7 available in GitHub Copilot?
Yes. GitHub added it on September 21, 2026 for Copilot Pro, Pro+, Max, Business and Enterprise, with a gradual rollout and usage-based billing at the provider's list price.
Is Grok 4.7 better than Claude or GPT?
On xAI's own published scores it is competitive but trails Claude Fable 5.1, for example 1695 vs 1735 on GDPval. Test it on your own tasks before switching.
What is Grok 4.7 designed for?
xAI positions it for agentic coding and complex, multi-step workflows. xAI says it reasons longer on hard problems and checks its answers more often than Grok 4.6.