Dead Release Among the two models of artificial intelligence (AI) Lama 4, the United States was placed as a leader in the artificial intelligence race, David SachsThe United States said of artificial intelligence and Crypto Czar, at X mail on saturday. books:
“In order for the United States to win the artificial intelligence race, we must win the open source as well, and return Llama 4 at the forefront.”
On April 6, Dead It has announced the launch of the Llama 4 Scout and Llama 4 MAVERICK.
The artificial intelligence race intensifies with the launch of Deepseek
Dibsic, a Chinese emerging company of artificial intelligence, was established in 2023, its first model in December 2024. In January 2025, she launched a chat claimed to be a ChatGPT capacity from Openai.
Deepseek downloads from artificial intelligence, Deepseek R1, applications charts. The main assumption in the world has shook the world: huge investments and expensive chips are the only way to go ahead in the artificial intelligence game.
As a result, the prices of American technology companies such as NVIDIA suffer Losses after the release of Deepseek R1.
This is because Dibsic claimed that he spent nearly $ 6 million to train the artificial intelligence model. On the other hand, Openai It is said It spent about $ 100 million to train ChatGPT-4.
While capitalism adventure Marx Andres It was called “The moment of Sputnik” from Dibsic, “The moment of Sputnik”, called by US President Donald Ji Trump.Waking up callingFor American companies.
Since then, American companies have been trying to jump in the artificial intelligence race, and according to the Sacks, who was an audio supporter of AI, Llama 4 was the key.
Meta Llama 4 allegations are “the best in their chapter”
Meta claims that Llama 4 models are “the most advanced models so far” and also “the best in its chapter for pluralism.” Both models are currently available for download and use on Meta applications such as WhatsApp and Instagram.
Multimedia artificial intelligence Systems are able to process different types of data – text, image, sound and video – simultaneously. This enables artificial intelligence to understand complex scenarios and generate comprehensive responses.
Llama 4 Scout and Llama 4 MAVERICK is the first open -source Amnesty International Mettha designed using a mix of architecture of experts (MEE). In MEE, smaller models or specialized experts are cooperating to make the most artificial intelligence model. This means that experts focus on solving the parts of the problem that are designed to deal with.
Llama 4 Scout has 17 billion active teachers and 16 experts. Llama 4 MAVERICK has the same number of parameters but designed with 128 experts. While the first can suit the one NVIDIA H100 graphics processing unit, the latter requires the H100 host.
Meta claims that Llama 4 Scout outperforms Gemma 3, Gemini 2.0 Flash-Lite and Mistral 3.1 through a wide range of widely reported standards.
Llama 4 MAVERICK, on the other hand, provides comparable results with Deepseek V3 on thinking and coding despite the presence of less than half of the active parameters. Meta also confirms that Llama 4 MAVERICK surpasses GPT-4O and Gemini 2.0 Flash via a number of criteria.
Moreover, according to the Meta test, Llama 4 “responds to strong political leanness at a similar rate of GROK.”
Meta also unveiled Lama 4 Beheemoth, which is still in training, as one of the world’s large language models (LLMS).
Dead Firing Lama’s first model in February 2023.
Mentioned in this article
