Meta Platforms (META, Financial) is reportedly getting ready to release the next version of its large language model, Llama 4, before the end of the month, according to The Information.
The launch has already been pushed back a couple of times due to performance issues, especially in reasoning and math. People familiar with the project say the model still trails behind OpenAI’s in terms of handling voice interactions that feel more human.
Meta rolled out Llama 3.3 in December, but the AI landscape has been moving fast since then. Google (GOOG, Financial), for example, launched Gemini 2.5 back in March.
Alongside the new model, Meta is working on something called Llama X — an internal effort to create a Llama API for businesses. The goal is to make it easier for companies to plug Llama into their own software. It could also open up a new revenue stream for Meta, according to sources.
Back in March, Morgan Stanley said it expected Llama 4 to show up sometime this spring. Meanwhile, Meta is gearing up for its developer event, LlamaCon, next week, where more updates are expected on its AI lineup.
Last fall, CEO Mark Zuckerberg said the upcoming models are being trained on more than 100,000 of Nvidia’s (NVDA, Financial) H100 chips — one of the largest setups seen so far. Meta also recently shared that its Llama models have been downloaded over one billion times.
So far, Meta hasn’t confirmed the exact timing for the Llama 4 launch.
