Meta
Latest in Meta
Meta’s open-source SAM 3 and DINOv3 models power SYNAPS-I, a US Department of Energy Genesis Mission project led by Lawrence Berkeley National Laboratory.
Meta Superintelligence Labs has launched Muse Image, an agentic image model ranked No. 2 on Arena, and previewed its Muse Video model.
Meta’s Brain2Qwerty v2 decodes brain activity into text without surgery, reaching 61% word accuracy from non-invasive MEG recordings.
The Commission has stepped in to stop Meta using WhatsApp as a moat against AI rivals, ordering free third-party access to be restored immediately.
Meta’s updated WhatsApp Business API terms ban general-purpose AI chatbots from external developers from 15 January 2026, affecting Copilot and ChatGPT.
Meta has released V-JEPA 2, a 1.2bn-parameter video world model for zero-shot robot planning, with three new physical reasoning benchmarks.
Meta has released Llama Guard 4, LlamaFirewall and Prompt Guard 2 for developers, plus CyberSecEval 4 and a Llama Defenders Program for security teams.
Litmos is using Meta’s Llama 3.1 8B to power an AI assistant that replaces keyword search with natural-language discovery across its corporate learning platform.
Meta’s FAIR team has released Perception Encoder, Perception Language Model, Locate 3D, Dynamic Byte Latent Transformer and Collaborative Reasoner.
AI video company Tavus says Meta’s Llama 3.3 models power its conversational video interface, helping developers build real-time digital twin experiences.
Digital advertising platform Smartly says Meta’s open-source Llama 3 now automates 80% of support ticket handling, freeing agents for complex problems.
Brazilian healthcare AI company Sofya says Llama models with knowledge distillation have cut provider documentation time by 30% for clinicians.
Orakl Oncology reports 26.8% better cancer treatment prediction with Meta’s open models, and Mendel AI cut clinical trial matching from hundreds of days to one.
Meta’s FAIR lab and the Basque Center on Cognition, Brain and Language decoded up to 80% of characters from non-invasive brain recordings.
Court docs claim Zuckerberg approved LibGen use for Llama training. Engineers stripped copyrights and used torrents despite legal concerns.