Meta’s answer to DeepSeek is here: Llama 4 launches with long context Scout and Maverick models, and 2T parameter Behemoth on the way!
1 min read
Summary
Meta has launched its latest AI model, Llama 4, in an attempt to keep up with DeepSeek and OpenAI.
The Chinese startup DeepSeek’s open-source language reasoning model DeepSeek R1 has outperformed solutions from Meta and other US companies, causing Meta to panic.
DeepSeek’s model was trained for a fraction of the cost but outperformed many other well-known models.
As a result of this, Meta has released its new multi-modal Llama 4 series of models, which are available to download for developers.
There are three models in the series, all of which have long context windows and are capable of text, video and imagery.
Meta’s CEO, Mark Zuckerberg, has said the company’s aim is to make “the world’s leading AI, open source it, and make it universally accessible”.
Although Meta’s blog post claims the new models are “more powerful than all previous generation Llama models”, it does not necessarily set new performance records.
It is, however, a powerful model family and is now a competitive open alternative to top-tier proprietary models from Google, DeepSeek and OpenAI.