Meta Llama 3.1 405B
Meta positioned Llama 3.1 405B as the world's largest and most capable openly available foundation model, and the first open model to rival closed frontier systems — claiming it was competitive with GPT-4, GPT-4o, and Claude 3.5 Sonnet across general knowledge, math, tool use, and multilingual tasks on 150-plus benchmark datasets. The release expanded context to 128K tokens and added support across eight languages, alongside upgraded 8B and 70B models. Crucially, Meta changed its license to permit using Llama outputs to train and improve other models, explicitly enabling synthetic-data generation and distillation. The framing was strategic: "open source AI is the path forward," positioning openness against the closed labs.
It shipped the same day as open weights for download on llama.meta.com and Hugging Face, plus immediate availability across partner platforms including AWS Bedrock, Azure, Google Cloud, Databricks, and NVIDIA, with hosted inference and fine-tuning options. The 405B model was released under the updated Llama 3.1 Community License rather than a fully OSI-open license.
The launch delivered on its headline claim — a genuinely frontier-scale model anyone could download — and reset the ceiling for what "open" AI could do. In practice, the 405B model itself saw limited direct deployment because running it requires substantial GPU infrastructure; its larger impact was as a teacher model whose newly permitted distillation rights fed a wave of smaller, cheaper open models. The license, while permissive, is not true open source (it carries a 700M-monthly-active-user commercial restriction), a nuance often lost in "open source" coverage. Strategically it validated the open-weights ecosystem as a real alternative to closed APIs for organizations prioritizing control, on-prem deployment, and data residency — even if most adopters used the smaller variants rather than the 405B flagship.
It gave organizations a credible frontier-grade option they can self-host and control rather than rent through an API, making "build on open weights vs. buy closed access" a real strategic choice. For executives, it is the anchor for any conversation about data sovereignty, vendor lock-in, and the open-vs-closed AI debate.