Back to AI Timeline
2023Model

Meta Llama 2 (Open Source)

Meta released Llama 2 as an open-source large language model available for both research and commercial use. With models ranging from 7 billion to 70 billion parameters, it gave developers and companies a powerful alternative to proprietary APIs. The release accelerated the open-source AI movement and enabled a thriving ecosystem of fine-tuned models.

In July 2023, Meta released Llama 2 as an open-source large language model available for both research and commercial use, fundamentally shifting the dynamics of the AI industry. With models at 7 billion, 13 billion, and 70 billion parameters, Llama 2 provided organizations of all sizes with a powerful, customizable alternative to proprietary APIs from OpenAI and others.

The Strategic Decision

Meta's decision to open-source Llama 2 was both philosophical and strategic. Mark Zuckerberg argued that open-source development leads to better, safer, and more innovative technology. Strategically, Meta stood to benefit from a world where AI capabilities were widely distributed rather than concentrated in the hands of competitors who controlled API access. By giving away the model, Meta could commoditize the layer of the stack where its rivals (particularly OpenAI and Google) were trying to build competitive moats.

The Model

Llama 2 was trained on two trillion tokens of publicly available data, roughly 40 percent more than the original Llama. It used a standard Transformer decoder architecture with several improvements including grouped-query attention for faster inference. Meta also released Llama 2 Chat, a version fine-tuned for dialogue using both supervised fine-tuning and RLHF, directly competing with ChatGPT-style assistants.

The License

Llama 2 was released under a custom license that was open but not fully permissive. It allowed commercial use for companies with fewer than 700 million monthly active users -- a threshold that excluded only the largest tech companies (effectively Meta's direct competitors). This approach balanced openness with competitive strategy, though some open-source purists criticized it for not being truly open.

The Ecosystem Effect

Llama 2's release triggered an explosion of derivative models and applications. The open-source community fine-tuned Llama 2 for specific domains (medical, legal, financial), languages, and use cases. Quantized versions that could run on consumer laptops appeared within days. Tools like llama.cpp made it possible to run the model locally on a MacBook. The ecosystem grew at a pace that no proprietary model could match.

Performance

Llama 2 70B was competitive with GPT-3.5 on many benchmarks, though it trailed GPT-4 significantly. The Llama 2 Chat variant performed well in conversational evaluations, and Meta claimed it was comparable to ChatGPT on helpfulness and safety. For many practical applications, Llama 2 was "good enough" -- and the ability to run it locally, customize it, and avoid per-token API costs made it attractive despite the capability gap with frontier models.

Impact on the Industry

Llama 2 changed the competitive dynamics of the AI industry. Companies that had been dependent on OpenAI's API suddenly had a viable alternative. Startups could build products on customized open-source models rather than being subject to API pricing changes or terms of service modifications. Enterprise customers with data privacy concerns could run models entirely on their own infrastructure.

The Open Source AI Movement

Llama 2 was the most significant catalyst for the open-source AI movement since Stable Diffusion. It demonstrated that open-source models could be competitive with proprietary offerings and that a major tech company would invest in creating and sharing them. The release inspired other companies -- including Mistral, Technology Innovation Institute, and others -- to release their own open models, creating a virtuous cycle of open AI development.

Legacy

Llama 2 established the template for how major companies could release open-source AI models for strategic advantage. Meta continued this approach with Llama 3 in 2024, which narrowed the gap with frontier models even further. The Llama series demonstrated that the future of AI would not be dominated solely by proprietary models behind APIs, but would include a robust ecosystem of open, customizable alternatives.

Key Figures

Mark ZuckerbergYann LeCunHugo Touvron

Lasting Impact

Llama 2 democratized access to powerful language models for commercial use, catalyzing the open-source AI movement and fundamentally changing the competitive dynamics of the AI industry. It enabled thousands of organizations to build AI products without dependence on proprietary APIs.

Related Events

2020Model
GPT-3 Launches

OpenAI released GPT-3 with 175 billion parameters, demonstrating remarkable few-shot learning abilities across a wide range of tasks without task-specific fine-tuning. Users could prompt the model with just a few examples and get high-quality outputs for translation, code generation, creative writing, and more. GPT-3 ignited a wave of AI startups and applications built on large language models.

2022Model
Stable Diffusion Goes Open Source

Stability AI released Stable Diffusion as an open-source image generation model, democratizing access to high-quality AI art creation. Unlike proprietary alternatives, anyone could download, run, and modify the model on consumer hardware. The release sparked an explosion of creative applications, fine-tuned models, and community-driven innovation.

2025Milestone
Open Source Models Close the Gap

Open-source language models from Meta, Mistral, and others reached performance levels competitive with proprietary frontier models across most practical tasks. The narrowing gap made powerful AI accessible to organizations of all sizes and reduced dependency on a handful of API providers. This democratization accelerated AI adoption across industries worldwide.