Elon Musk says xAI finished training its 1.5-trillion-parameter Grok V9-Medium model using Cursor data for supplementary training
Public release is scheduled within two to three weeks.
@eliebakouch We will open source the 0.5T model towards the end of this year. It should still be quite useful.
elie@eliebakouch@elonmusk congrats! looking forward for the release :) any plan on open sourcing the 0.5T model? would be super cool
#167Emad@EMOSTAQUEIt’ll be interesting to see if the post training for this uses a multiple of the compute of pretraining as cursor did when they tuned Kimi as the base model
Elon Musk@elonmuskGrok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come. Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release. This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
@elonmusk Excited to see
Elon Musk@elonmuskGrok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come. Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release. This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
#341Bill Yuchen Lin@BILLYUCHENLINGrok Build being built in public 🚀
Elon Musk@elonmuskGrok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come. Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release. This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
> A lot of Cursor data was added in supplementary training and there is more to come. ok, maybe xAI is getting serious
Elon Musk@elonmuskGrok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come. Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release. This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
Can anyone explain why Elon has such a slow opensourcing cycle? The cynical thesis for Chinese labs is "they're behind, the competition can't use it to accelerate, & they can't serve all demand anyway". Well, Grok 4.x is behind even China, and unlike them, Elon can serve it!
Elon Musk@elonmusk@eliebakouch We will open source the 0.5T model towards the end of this year. It should still be quite useful.
like this is ridiculous. Elon still hasn't open sourced Grok fucking 3, and likely never will at this point.
> end of year we gonna have open source models that are better than ANY frontier model rn including opus 4.7 and gpt 5.5 Realistically it'll be K3 or late stages of V4. I am *dubious* you can recreate 5.5 with that. It is big and it knows a lot. It's a scary good model.
Maybe V4 with a lot of continued pretraining, a la Kimi K2 to K2.5. I can see the stacking of tricks that would amount to this result, but it's on the margins of possibility. 5.5 in its generality, not on select near-saturated benchmarks, is not a trivial milestone.
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex> end of year we gonna have open source models that are better than ANY frontier model rn including opus 4.7 and gpt 5.5 Realistically it'll be K3 or late stages of V4. I am *dubious* you can recreate 5.5 with that. It is big and it knows a lot. It's a scary good model.
#716elie@ELIEBAKOUCHonly 2-3 weeks for the RL phase of a 1.5T model is an interesting data point. also interesting that he mentioned cursor data was added before post training, they're likely doing heavy mid training like in composer 2.5 and wilkl focus heavily on code
Elon Musk@elonmuskGrok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come. Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release. This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
#716elie@ELIEBAKOUCH@elonmusk congrats! looking forward for the release :) any plan on open sourcing the 0.5T model? would be super cool
Elon Musk@elonmuskGrok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come. Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release. This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
#716elie@ELIEBAKOUCH@elonmusk huge!!
Elon Musk@elonmusk@eliebakouch We will open source the 0.5T model towards the end of this year. It should still be quite useful.
#839Beff (e/acc)@BEFFJEZOSGrok Build will finally have a solid model to go with the great frontend
Elon Musk@elonmuskGrok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come. Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release. This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
@elonmusk let's goooo
This will make Grok build a pleasure to use
Elon Musk@elonmuskGrok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come. Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release. This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
Wow, Grok v9-Medium - a 1.5T model will be launching in June
Will be FANTASTIC to have Grok join the the top 3 model labs
Fundamentally, the real-time X data gives them a huge advantage in the long run