- Goal has launched three new calls 4 llms
- Download call 4 scout and maverick from flame.com or hugging face today
- You can try call 4 right now on WhatsApp, Messenger and Instagram Direct
Meta has launched what is calling a new “herd” of flame models 4. There are three flavors of the new flame 4, called Scout, Maverick and Behemoth, and two are available at this time to try it on their goal applications such as Messenger, Instagram and WhatsApp.
Call 4 is the latest flagship version of Meta’s Open Source calls AI, and the new launch occurs almost exactly a year after the launch of flame 3 in 2024.
Inspired by training advances made by Deepseek, the new flame 4 has been trained using the most efficient mixture of the ‘expert’ methodology.
As the names suggest, Scout is the lightest model, with 109 billion parameters, while Maverick has 400 million parameters. Both models are available at this time for developers to discharge it, and are also used in the popular Meta Consumer applications.
Heavy
Call 4 Behemoth is a teacher centered, which Meta claims to perform GPT-4.5, Claude, Sonnet 3.7 and Gemini 2.0 Pro at reference points centered on Stem such as Math-500 and GPQA Diamond. There is currently no access to flame 4 Behemoth, since Meta says he is still “in training.”
The new goal models keep it at the forefront of Competitive Open Source. While the reference points are impressive, the current consumer experience of using a goal AI still is left using chatgpt or gemini.
For example, while both AIS DE flame 4 available are multimodals, there is still no way to load an image through the goal, or in one of the many target applications. You can ask goal to look at the UR of an image and analyze what is seen, but the direct load is not compatible.
Likewise, goal AI lacks other extra chatbot extras that we have thought of as standard these days, such as the search for AI and deep reasoning, and their image generation capabilities are delayed behind the most recent chatpt update.
Copyright problems remain
The new flame 4 models are accessible to developers, which can download open source models to use competitive token rates on flame.com and embrace the face.
Alternatively, the new flame 4 LLM are available at this time to use in goal.AI or in the target applications such as Messenger, WhatsApp and Instagram Direct.
It is worth noting that the new flame 4 LLMS is still part of a current copyright dispute between the goal and several famous authors after judicial documents alleged that the CEO of Meta Zuckerberg had approved the use of the set of data from Libgen, among other shadow libraries, in the training of its LLM flame.
The Atlantic recently published a database to search titles contained in Libgen, which allows many authors to see if Meta could have been training their AI in their work without permission.