Meta has announced the a la mode looping of its open - reference AI exemplar family Llama 4 , which the mark has evolve while competition in the generative AI manufacture continue to intensify .
The new AI familyincludes four mannequin , and Meta detail Llama 4 Scout , Llama 4 Maverick , and Llama 4 Behemoth . Meta detail on its AI internet site that the example were train on “ large amounts of unlabeled text , image , and television datum . ” This indicates that the models will have varied multimodal capabilities .
Currently , two of the models – Llama 4 Scout and Llama 4 Maverick- are available for user to access across Meta ’s platforms , include WhatsApp , Messenger , and Instagram Direct , in gain to Meta ’s AI website , Llama.com , as of Saturday . The developer can also get at AI models at unresolved - source repositories such as Hugging Face . The Llama 4 Behemoth good example is not yet turn and is still in education . The company has indicated that the industry should expect the Behemoth model to outperform corresponding models , and that it will serve as a instructor for the other models in the Llama 4 kin .
Amid its interior examination , Meta compared the Llama 4 models to competitor AI applied science to determine their capacity and good use cases . The company indicated that Llama 4 Maverick would work best for originative committal to writing . to boot , it outperformed the OpenAI GPT-4o and Google Gemini 2.0 example in role such as coding , logical thinking , multilingual , long - linguistic context , and image genesis . Meanwhile , Maverick clamber to keep up with the capabilities of AI model , includingGemini 2.5 Pro , GPT-4.5 , andAnthropicClaude 3.7 Sonnet .
While Meta arrogate that Behemoth can outperform most of these models , except for Gemini 2.5 Pro , the caller has n’t been capable to reduce the hardware cost of training its most powerful model .
TechCrunchnoted that the tending the Chinese AI companyDeepSeekhas gained from its free-enterprise yet inexpensive models put Meta on notice . The ship’s company has reportedly been intensely studying how the rival company developed its notable models , let in R1 and V3 , at dispirited operating toll thanprior Llama exemplar .
The company detail that the Llama 4 Scout model can move on oneNvidiaH100 GPU . The Llama 4 Maverick modeling can go on one Nvidia H100 DGX graphics organisation .
Meta is set to host its firstLlamaConAI conference on April 29 . The company also has a standalone Meta AI chatbot that is set to launch in the second tail of the twelvemonth , according toCNBC .
Meta is n’t the only company being powdered ginger with the timeline of its major AI models . OpenAI recently adjusted the launch of itsGPT-5 model , with the company ’s CEO , Sam Altman , announcing on social medium that rooter should expect novel o3 and o4 - mini logical thinking models in the come weeks as an choice to GPT-5 . The executive detailed that GPT-5 will now launch in the coming months , which will give OpenAI additional time to get the model up to standard .
