Elon Musk posted on X early Monday morning (5:48 AM UTC), May 25, that xAI has finished training Grok foundation model V9-Medium. According to him, this model will come with 1.5 trillion parameters, which is three times bigger than the current version running Grok today.
The public release timeline is between 2 and 3 weeks, with Musk saying evaluations look good, fine-tuning is underway, and reinforcement learning starts in a few days.
What caught attention, though, is Musk's focus on coding. When asked if the new model would perform better at coding tasks, Musk replied directly: "Much better at coding."
Damilare Odedina
What V9-Medium means and why Cursor data matters
The V9-Medium name refers to xAI's internal model version, not the public product name users see. The current Grok app runs on V8, which has 0.5 trillion parameters. V9-Medium triples that size to 1.5 trillion parameters.
“This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic,” Musk said in his post.
Parameter count is basically the number of connections inside an AI model. More parameters typically mean the model can handle more complex tasks, though size alone doesn't guarantee better performance.
Musk revealed that xAI added "a lot of Cursor data" during training, with more still coming. Cursor is a code editor that developers at OpenAI, Stripe, and Perplexity actually use. It's basically VS Code with AI features built in for writing and debugging code.
Training on Cursor means Grok learned from how real developers work, not just public code on GitHub.
Where Grok stands against Claude and ChatGPT
Claude leads on coding right now. Independent testing tests conducted by Ryz Labs shows it hits about 95% accuracy on coding tasks. ChatGPT comes in around 85%. Claude's Opus 4.6 scores 80.8% on SWE-bench Verified, which developers watch closely.
GPT-5.5 scored 88.7% on that same test, while xAI self-reports Grok 4, it’s current flagship series of models, at 72% to 75% . So there's ground to make up.
If V9-Medium closes the performance gap, the coding AI market gets more competitive. The Cursor training data suggests xAI knows where Claude is strong and is aiming there.
Musk also said xAI will open source the 0.5 trillion parameter model "towards the end of this year." Developers could use that to experiment while xAI keeps pushing bigger models.
Mid-June 2026 is when V9-Medium should arrive, based on the 2 to 3 week timeline. Whether it matches Claude and ChatGPT on actual coding benchmarks remains to be seen.
Damilare Odedina
