CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins84% CONFIDENCEOverall confidence: 84%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better
The announcement is dated "Sep 21, 2026" and says "Grok 4.7 is available today in Cursor and Grok Build"; the xAI API release notes list it under September 21, and Decrypt[4] says "xAI released Grok 4.7 on Monday" (21 September).- Start
- Sep 21, 202690% CONFIDENCEsource
Grok 4.7 Released
SpaceXAI, formerly xAI, releases Grok 4.7, its most capable model for coding and knowledge work, built on a larger base model and priced like Grok 4.6 at $2 per million input tokens; independent tests still put it well behind Claude Fable 5.1 and GPT-6.
- SpaceXAI released Grok 4.7 on 21 September 2026, calling it "our most capable model for coding and knowledge work", served "at the same price and speed as Grok 4.6"[1]
- It uses a new, larger base model, a longer reinforcement-learning run weighted toward problems that take many hours, better self-verification and long-context management, and was trained to work natively in the Grok Bot harness[1]
- Decrypt reports it has 2.1 trillion parameters, up 40% from Grok 4.6's 1.5 trillion, with supplemental training data from SpaceX: Starlink satellite telemetry, manufacturing records and engineering failure logs[4]
- It came after Elon Musk walked the timeline back at least five times since late July, from "four weeks out" to "needs a few more days to cook" on 11 September; days before launch he said it should land "roughly on par with" Claude Opus 5.0[4]
- Independent tests put it mid-pack: 46 on the Artificial Analysis Intelligence Index against 53 for Claude Fable 5.1 and GPT-6, and 26% on Artificial Analysis' Terminal-Bench 4.0 run against 60% for GPT-6 Astra[5]
- It went live in the Grok app, Cursor, Grok Build and the API, and GitHub began rolling it out in Copilot the same day[1][4][6]
Notable features
- Prices per million tokens: $2 input, $0.50 cached input and $6 output below 200k prompt tokens, and $4, $1 and $12 above; a Grok 4.7 Fast variant with twice the output speed costs twice as much and is offered only in Cursor and Grok Build[1][2]
- A 500k-token context window, text and image input with text-only output, no output limit, and reasoning effort from low to xhigh, with high the default[2]
- SpaceXAI's own scores (xhigh): CursorBench 4.0 46.3% against 40.4% for Grok 4.6 and 51.8% for Fable 5.1, DeepSWE v1.1 71.0% (high effort), EEBench 64.0%, Terminal-Bench 4.0 37.6% and GDPval 1695 Elo against Fable 5.1's 1735[1]
- An entirely new safeguard stack: it tops LatchBio's biosafety benchmark at 62.4% and lets only 3.3% of risky dual-use prompts through on SpaceXAI's HackerBench v0.3[1]
- Select cybersecurity partners get invite-only access to its red-team capabilities for defence research[1]
Training
SourceParameters
2.1T
model weights
Training compute
Unknown
not published
Training hardware
Unknown
not published
Training time
Unknown
not published
Training dataset size
Unknown
not published
Buy onGrok