Grok 4.7 release date slipped — stay on 4.6 for now
Grok 4.7 release date is unconfirmed as of Sep 13, 2026. Stay on Grok 4.6 at $2/$6 per million; a $2k–$4k Telegram agent still owns the console.
There is no Grok 4.7 release date. As of September 13, 2026, xAI has not published a model ID, a price, or a launch post. Grok 4.6 is still the model you can actually call.
Owners freeze deploys because a bigger version number is “any day now.” That is a lab calendar. Missed calls and CRM notes do not wait on a model that is not in the docs.
What is the Grok 4.7 release date right now?
There isn’t one. xAI’s September 2026 developer release notes still list Grok 4.6 as the current frontier model. There is no grok-4.7 slug, no token price, and no model card. Treat every circulating date as unofficial until that page changes.
Grok 4.6 shipped in August 2026. Official notes give it a 500k context window, text and image inputs with text-only output, and no text output limit. API pricing is $2 input / $0.50 cached / $6 output per million tokens below 200k prompt tokens, and $4 / $1 / $12 above. Reasoning effort supports low, medium, high (default), and xhigh.
That is the stack you can pin today. A September 12 window that had been circulating on X did not produce a ship. Public comments the day before pointed at more reinforcement-learning work and no replacement date. I am not going to pretend a tweet is a changelog.
If you already consume Grok through Microsoft Foundry, this delay does not change the Grok 4.6 Microsoft Foundry listing. Foundry is still a model endpoint, not the workflow.
Does the delay change a Grok Bot or custom agent deploy?
No. Grok 4.7 is a model bump. Grok Bot is a named teammate with a shared cloud computer. A Telegram agent is a locked owner console you own after a one-time build. Waiting on 4.7 does not fill any of those boxes.
The path I still deploy is the same one I use for AI for small business:
- Trigger: a Telegram message, a voice note, a form, or a Grok Bot routine.
- Agent action: read it, structure the fields, draft the next step.
- System of record: HubSpot, Jobber, Housecall Pro, or a shared sheet — not the chat thread.
- Human escalation: you, for quotes, refunds, deletions, and anything that spends money.
Grok 4.6 can sit in step 2. When 4.7 actually lands, you swap the model ID and re-run the same evals. You do not rebuild the trigger, the CRM write, or the kill switch.
Grok Bot vs a custom agent is the real fork. SuperGrok Heavy, Cursor Ultra, or Cursor Teams Premium already include Grok Bot for overnight research. If you need the bot on a phone chat with your keys, that is still the $2,000–$4,000 Telegram AI Agent. Neither one is “wait until 4.7.”
Grok 4.7 will not answer the business phone. If the leak is missed calls, that is a receptionist problem, not a model-version problem.
What should you run until Grok 4.7 actually ships?
Pin Grok 4.6, cap the API key, and keep the owner console you already decided on. Do not open a second chat surface “for the new model.” Do not pause after-hours coverage because a rumor said Friday.
Practical checklist:
- Lock the model string to
grok-4.6(or the Foundry equivalent) in whatever you already run. - Put a daily spend cap on the key. A 500k context window is a gift and a bill if a loop retries a CRM write.
- Keep one system of record. Telegram is the console. HubSpot or Jobber is truth.
- Write the swap runbook now: change model ID, replay last week’s ten real jobs, compare CRM notes, then promote. That is a Tuesday afternoon, not a rebuild.
Skip the wait if you do not have a Grok key. Then the first job is still intake, notes, and follow-up.
When xAI actually posts a grok-4.7 ID and a price, I will treat that as the announcement. Until then, run 4.6. If you want the owned console instead of another chat tab, I hand-deploy the Telegram agent for $2,000–$4,000 one time. You own the setup. Usage stays on your xAI (or Cursor) bill.
