Meta Llama 2
Meta positioned Llama 2 as a credible, freely available alternative to closed frontier models, releasing it "free for research and commercial use" in partnership with Microsoft as preferred cloud partner. It came in three sizes (7B, 13B, and 70B parameters) with both pretrained base models and chat-tuned variants, and Meta emphasized responsible-release practices: red-teaming, a responsible use guide, and an acceptable use policy. The framing was explicitly open: democratizing access to capable LLMs that organizations could self-host and fine-tune.
Open weights and starting code shipped the same day across Azure AI, AWS, and Hugging Face, with optimization to run locally on Windows. Use was free but governed by Meta's custom "Llama 2 Community License" rather than an OSI-approved open-source license.
Llama 2 was the launch that made open-weight models a serious enterprise option, seeding a vast ecosystem of fine-tunes, hosted endpoints, and on-prem deployments that gave organizations a path off the API-only frontier vendors. The chat models trailed GPT-4-class quality at launch, but for many workloads the 70B model was good enough and the self-hostable, fine-tunable footprint mattered more than topping a leaderboard. The "open" label was contested: the license blocks firms with over 700 million monthly active users without Meta's permission and bars using outputs to improve competing LLMs, so it is best described as "open weights," not open source. Strategically it was Meta's commoditize-your-complement play, applying downward pressure on closed-model pricing and establishing Llama as the default base for the open-model world.
It established a real, deployable alternative to closed APIs, giving executives leverage on cost, data control, and vendor lock-in, but the restrictive license means "open" should be read carefully rather than assumed.