The Drum · Weekly AI news, freeSubscribe →
Timeline
Meta's Llama models represent a watershed moment in artificial intelligence history. Released initially in February 2023, Llama brought large language model capabilities to researchers and developers outside the wealthiest tech companies. The open-source approach democratized access to frontier AI systems, sparking a global movement that challenged the closed-model dominance of OpenAI and Google. By 2024, Llama had become the foundation for thousands of derivative projects, fine-tuned variants, and commercial applications. This shift fundamentally altered how AI research progresses, who can participate in it, and what governance frameworks might emerge for powerful AI systems.
The Story
The story begins with Meta's decision to release Llama weights to researchers in early 2023, a move that seemed counterintuitive for a company competing with OpenAI and Google. Within weeks, the model leaked publicly, forcing Meta to acknowledge the shift and eventually embrace open-source distribution. What followed was explosive growth: Llama 2 arrived with commercial licensing, spurring a wave of fine-tuned variants like Alpaca, Mistral, and countless others. By 2024, Llama had become the de facto foundation model for open-source AI, with major cloud providers integrating it and enterprises building production systems on top. The unresolved tension remains: can open-source AI coexist with safety standards and corporate liability, or will regulatory pressure force a recentralization of model access?
2023-02
Meta releases Llama model weights to researchers
Meta distributed Llama model weights to academic researchers and select institutions, marking a significant departure from the closed-model approach of competitors. The move aimed to accelerate AI research while maintaining some control over distribution. Within days, the weights leaked to public repositories, forcing Meta to confront the reality of open-source AI.
2023-03
Llama weights circulate widely on GitHub and torrents
Despite Meta's initial restrictions, Llama model weights spread rapidly across public platforms. Researchers and developers worldwide gained access to the 7B, 13B, and 65B parameter versions. This uncontrolled distribution accelerated the open-source AI movement beyond Meta's original scope.
2023-03
Stanford releases Alpaca, first major Llama fine-tune
Researchers at Stanford created Alpaca by fine-tuning Llama on instruction-following data. The project demonstrated that smaller, open-source models could match or exceed proprietary systems in specific tasks. Alpaca became a template for hundreds of subsequent fine-tuning efforts.
2023-04
Meta acknowledges Llama leak, shifts to open distribution
Rather than pursue legal action, Meta formally acknowledged the leak and began distributing Llama through official channels. This pragmatic pivot signaled Meta's commitment to the open-source model and positioned the company as a leader in democratized AI.
2023-05
Mistral AI founded by former Meta researchers
A team of researchers, including alumni from Meta's AI division, launched Mistral AI with a focus on efficient, open-source language models. The company would become one of the most prominent players in the open-source ecosystem.
2023-06
Llama-based models exceed 100 public variants
By mid-2023, over 100 distinct fine-tuned versions of Llama existed on Hugging Face and other repositories. Variants targeted specific domains: medical, legal, coding, multilingual, and more. This explosion demonstrated the model's versatility and the community's creativity.
2023-07
Llama 2 released with commercial license
Meta released Llama 2, a significantly improved version, with explicit commercial licensing terms. The move legitimized business use of open-source Llama and attracted enterprise adoption. Llama 2 became the foundation for countless production systems.
2023-08
AWS, Google Cloud, and Azure integrate Llama 2
Major cloud providers began offering Llama 2 through their platforms, making deployment easier for enterprises. This integration signaled mainstream acceptance of open-source AI in production environments. Pricing models emerged for inference and fine-tuning services.
2023-09
Mistral releases 7B model, challenges Llama dominance
Mistral AI released its 7B parameter model, which matched or exceeded Llama 2's performance on many benchmarks. The release sparked healthy competition within the open-source ecosystem and proved that multiple strong alternatives could coexist.
2023-10
Open-source models reach parity with GPT-3.5 on some tasks
Benchmarking studies showed that fine-tuned Llama variants could match or exceed GPT-3.5 performance on specific domains. This milestone challenged the narrative that only frontier labs could build competitive AI systems.
2023-11
Llama 2 Chat variant released for conversational AI
Meta released Llama 2 Chat, optimized for dialogue and instruction-following tasks. The variant became popular for building chatbots and conversational applications without relying on proprietary APIs.
2023-12
Open-source AI community reaches 50,000+ developers
Hugging Face reported that the open-source AI community had grown to include tens of thousands of active developers. Conferences, Discord communities, and GitHub discussions became vibrant hubs for collaboration and knowledge sharing.
2024-01
Mistral raises 105 million euros in Series A funding
Mistral AI secured significant venture funding, valuing the company at over 2 billion euros. The investment validated the commercial viability of open-source AI and attracted attention from other investors.
2024-02
Llama 2 surpasses 10 billion downloads
Meta announced that Llama 2 had been downloaded over 10 billion times across various platforms. The milestone underscored the model's dominance in the open-source landscape and its adoption at scale.
2024-03
Concerns emerge about open-source model misuse
Security researchers and policy experts raised concerns about potential misuse of open-source models for generating harmful content or conducting cyberattacks. Debates intensified about balancing openness with safety and responsibility.
2024-04
Meta announces Llama 3 with improved capabilities
Meta released Llama 3, featuring enhanced reasoning, coding, and multilingual abilities. The model maintained Meta's commitment to open-source distribution while narrowing the gap with proprietary frontier models.
2024-05
EU AI Act raises questions about open-source compliance
As the EU AI Act took effect, regulators and industry experts debated how open-source models should comply with new requirements. Questions arose about liability, documentation, and enforcement when models are distributed globally.
2024-06
Llama-based models deployed in healthcare and finance
Enterprises in regulated industries began deploying fine-tuned Llama models for clinical decision support and financial analysis. These deployments demonstrated that open-source models could meet enterprise security and compliance standards.
2024-07
Open-source AI contributes to 30 percent of new AI startups
Industry analysis suggested that roughly 30 percent of newly funded AI startups were built on open-source foundations like Llama. The trend highlighted how open-source democratization was reshaping the startup ecosystem.
2024-08
Anthropic releases Claude model, maintains closed approach
Anthropic released Claude, a competitive frontier model, but chose not to open-source it. The decision highlighted ongoing strategic divergence between companies on the openness question.
2024-09
Llama 3.1 released with 405B parameter version
Meta released Llama 3.1, including a massive 405B parameter model available under open-source terms. The release demonstrated Meta's continued commitment to scale and openness simultaneously.
2024-10
Open-source models account for 40 percent of inference workloads
Cloud provider data suggested that open-source models were handling roughly 40 percent of LLM inference workloads. The shift reflected growing enterprise confidence in open-source alternatives.
2024-11
Debate intensifies over open-source AI governance
Policymakers, researchers, and industry leaders debated frameworks for governing open-source AI systems. Tensions emerged between innovation incentives and safety requirements, with no consensus yet reached.
2024-12
Llama ecosystem valued at over 50 billion dollars
Analysts estimated that the economic value created by Llama and derivative projects exceeded 50 billion dollars. The figure included direct commercial applications, research contributions, and startup valuations.
© 2026 Cedar & Bloom. All rights reserved.
