Meta Llama 4 Released: Native Multimodal + Long Context, Open-Source Model Stuns AI Community Again

Meta officially launched the Llama 4 series, including Llama 4 Scout (17B parameters, 16-expert MoE) and Llama 4 Maverick (400B parameters). Llama 4 natively supports image, video, and audio inputs, with a context window of up to 10 million tokens, surpassing Gemini 1.5 Pro on multiple multimodal benchmarks. More importantly, all models are commercially usable, and Meta announced deep integration of Llama 4 into WhatsApp and Instagram. The open-source community has seen over 300,000 downloads within 48 hours of release.