We are soon going to get a new 1.5T model by xAI
9:12 AM · May 15, 2026 · 11.4K Views
xAI completes training run for Grok V9 model
xAI completed an internal training run for its Grok V9 foundation model with 1.5 trillion parameters. The model is three times larger than the 0.5 trillion parameter base used in the current Grok 4.2 release. Initial results are strong ahead of supplemental training that will incorporate data from Cursor. Internal versioning marks Grok V9 as a distinct step from version 8 with major differences in scale and training approach.
2073.9K98306352.1K
Sentiment
Positive users welcome the Grok V9 1.5T model's training completion and expected capability gains plus added competition, while negative users object to privacy issues with repurposed Cursor data and question timeline accuracy.
Pos
70.2%
Neg
29.8%
34 comments with sentiment.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Posts from X
Most Activity
Most Activity
VIEWS302.7KBOOKMARKS255LIKES3.2KRETWEETS70REPLIES128
Theo - t3.gg@theo
Called it, they are gonna use Cursor’s data to leapfrog
Elon Musk@elonmusk
@beffjezos Our recently completed Grok V9 1.5T run is looking great and that is before Cursor data is added in supplemental training
45dViews 302.7KLikes 3.2KBookmarks 255
Chubby♨️@kimmonismus
According to Elon, Grok 4.2 is based on foundation model v8: 0.5T parameters, trained on Hoppers, with major data-quality shortcomings.
The new v9 model is 1.5T parameters, trained with a better recipe, better data curation, and optimized for Blackwell.
Better model with heat up competition even nore
45dViews 24.3KLikes 333Bookmarks 31
Rohan Paul@rohanpaul_ai
"The difference between (the current) Grok foundation model 8 and 9 is gigantic." ~ Elon Musk
Grok V9 is a 3x larger foundation model built to compete with top coding agents.
Elon Musk@elonmusk
The version numbers are a little confusing and deserve some explanation.
Internally, we are working on version 9 of our new foundation model, which is 1.5T params. This is substantially better in every way than v8: data curation, training recipe, size, etc. It is also optimized to run on a Blackwells.
The public facing v4.2 is based on foundation model v8, trained on Hoppers, with significant shortfalls in training data quality, comprehensiveness and proportionality. It is also only 0.5T in size.
The difference between Grok foundation model 8 and 9 is gigantic.
45dViews 7.3KLikes 57Bookmarks 2
Matthew Berman@MatthewBerman
@theo Think they’ll actually leapfrog?
Theo - t3.gg@theo
Called it, they are gonna use Cursor’s data to leapfrog
45dViews 6.4KLikes 25Bookmarks 2
Rhys@RhysSullivan
@theo somewhat concerned on xAI getting a really good coding model having bad cybersecurity implications given how little restrictions they put on their normal models
45dViews 2.3KLikes 42
Ezequiel@_zeke1
@theo It was obvious. It was literally announced in the partnership
45dViews 1.1KLikes 39
Jinjing Liang@JinjingLiang
@theo If that's true... Why isn't GitHub in the lead? Don't they have "all" the data and had co-pilot before anyone else? 🤔🤔
45dViews 1.8KLikes 9
maria@maria_rcks
@theo ...
45dViews 484Likes 11
m-ret@mretsal
@theo Don't you believe Cursor will become an absolute beast after this deal?
45dViews 2.7KLikes 1
Ananth@Ananthr27104587
@theo Think it was quite obvious and I really hope grok gets good
45dViews 1.6KLikes 9
ruminate - e/acc@empyrealrum
@JinjingLiang @theo Everyone scrapped the GitHub data. What is actually valuable is the back and forth between devs and the AI, cursor has all that data.
45dViews 287Likes 6
Paweł J Lisowski@PawelJLisowski
lets hope they do, the more competition the better for us! but they still need a lot optimisation, probably code data is not enough, real help might be some of talent cursor has since they were able to bring out the most from frontier models too. Either way looking forward to it.
45dViews 1.3KLikes 8
TravelerOfCode@TravelerOfCode
@theo training on cursor's data is the natural next move for anyone with the gpus and ambition. cursor was the moat 18 months ago. once the data leaks into training sets, everyone catches up. classic.
45dViews 1.4KLikes 7
building with no code@TFisPython
@theo training on usage data is how every dev tool eventually eats its integrations. Cursor built the moat, now the model shops use it to close the gap. seen this movie before with every API layer that got too popular.
44dViews 1.1KLikes 4
Jinjing Liang@JinjingLiang
@empyrealrum @theo But GitHub had VSCode's Copilot before Cursor existed
45dViews 271
Vismay Shrouty@BuildZaFuture
@kimmonismus Given the recent hermes-grok Integra and grokbuild-cli Idts grok will be far behind in the coming few months in terms of agentic coding The cursor deal is just a plus point to this.
45dViews 282Likes 3
Greg Val@val__greg
@theo users wrote code in Cursor under one privacy expectation. xAI repurposing it for training is a second move that nobody signed up for and can't be undone after it happens
44dViews 488Likes 1
Bennett@b_nnett
@RhysSullivan @theo this is just the world we live in now unfortunately. a lab who doesn't care will release a model really-capable of finding 0 days within the next 18 months.
45dViews 114Likes 2
elliott@cyburke
@theo I am here for all of the ass kicking. And I mean that for any player. I'm excited for pressure and how that leads to change, for all, and hopefully for the better. I've been talking about Grok for months and looking forward to seeing how they continue to grow. 👌
45dViews 2.5KLikes 3
aipulsedaily@aipulseda1ly
Is this a bad thing though? Elon does have the hardware to fast pace their performance and training. Really at the current moment in the leading space we have opus, gpt 5.5 and Gemini if we even consider Gemini as leading. Would not be so bad to have another frontier option in play.
45dViews 2.4KLikes 3
Load more posts