Elon Musk said Grok 4.7 would land around 12 September 2026, then pushed the date back. Here is what is actually confirmed, what is still just a claim, and what to do while you wait.
1. What is Grok 4.7?
Grok 4.7 is the model Elon Musk has said will follow Grok 4.6 as xAI's next flagship. As of 12 September 2026 it has not been released, and xAI's own developer docs list no Grok 4.7 model, pricing or context window.
Grok 4.6 became xAI's flagship in August 2026, with a published model card, a 500k token context window and API pricing. Grok 4.7 is being talked about as the model that replaces it, but talk and a shipped product are not the same thing.
2. Is Grok 4.7 out yet?
No. Checking xAI's developer model docs on 12 September 2026 shows Grok 4.6 still listed as the current flagship, with no Grok 4.7 entry, model identifier, price or context window published anywhere on xAI's official pages.
That matters because plenty of blog posts and social media threads describe Grok 4.7's specifications as though the model already exists. Until xAI publishes its own model card, the safer position is that Grok 4.7 is announced, not released.
Every specific number attached to Grok 4.7 so far comes from Elon Musk's own posts, not from an xAI model card.
3. When is Grok 4.7 coming out?
There is no confirmed release date. On 2 September 2026, Musk said on X that Grok 4.7 would arrive in ten days, pointing at around 12 September. On 11 September he said the model needed a few more days, which pushed the launch past that window with no new date given.
xAI has not set a replacement date at the time of writing. Treat any specific date you see elsewhere as someone's guess rather than a confirmed schedule, and check the official xAI news page and developer docs before relying on one.
4. Why was Grok 4.7 delayed?
Musk said the delay comes down to reinforcement learning tuning. He described a specific problem: the training process may have penalised long responses too heavily, so the model gives up on hard tasks it could actually solve, and it is not yet rigorous enough about checking its own work before answering.
That is a reasonable, fixable kind of problem rather than a sign the model failed outright. It also means the extra time is going into behaviour and reliability, not just raw scale.
5. How many parameters does Grok 4.7 have?
Musk has said Grok 4.7 is being built at around 2.1 trillion parameters, roughly 40% larger than the 1.5 trillion he has attributed to Grok 4.6. Both figures come from his own posts rather than an xAI technical report, so they should be treated as reported claims, not confirmed specifications.
xAI does not routinely publish exact parameter counts for its models, so there may never be an official figure to check these numbers against. If a precise parameter count matters to your decision, wait for it to appear in xAI's own documentation.
6. Grok 4.7 vs Grok 4.6
Grok 4.6 is the safer point of comparison because it is confirmed: released in August 2026, priced at USD 2 per 1 million input tokens and USD 6 per 1 million output tokens under 200k tokens, with a 500k token context window.
For Grok 4.7, Musk has claimed it will outperform Grok 4.6 on nearly every measure while running slightly slower, helped by a supplementary training pass on SpaceX engineering data. Those are claims from the model's own creator, not an independent benchmark, so treat any comparison as provisional until xAI or outside testers publish results.
7. How much will Grok 4.7 cost?
There is no official price for Grok 4.7, because the model has not shipped. xAI has not listed input or output token rates, and there is no equivalent yet to the pricing table it published for Grok 4.6.
For context, xAI kept Grok 4.6 at the same token price as Grok 4.5 when it launched, treating the update as a quality gain rather than a new pricing tier. That is a pattern worth watching, but it is not a promise about what Grok 4.7 will cost.
8. Should you wait for Grok 4.7?
For most people, no. Grok 4.6 is available now, confirmed and priced, while Grok 4.7 exists only as an announcement with a slipped date. Base a decision on what you can actually use today.
If you are building a product on the xAI API, design your integration so the model name is easy to swap, then revisit the choice once Grok 4.7 has a real model card, price and context window, rather than committing around numbers that are still just claims.