Meta had broken from the rest of the big-tech pack by choosing to open-source its AI models, but it seems that it’s walking back on that policy.
Meta has said that it will be “careful” in deciding which AI models it’ll choose to open-source. “We believe the benefits of superintelligence should be shared with the world as broadly as possible,” Meta announced today. “That said, superintelligence will raise novel safety concerns. We’ll need to be rigorous about mitigating these risks and careful about what we choose to open source. Still, we believe that building a free society requires that we aim to empower people as much as possible,” it added.
This seems to be a bit of a volte-face from what Meta had declared exactly a year ago. On 23rd July 2024, Meta CEO Mark Zuckerberg had written a post titled “Open Source AI is the Path Forward”. Zuckerberg had highlighted how open-source solutions like Unix and Linux had managed to gain prominence, and said that he believed that AI would go down a similar route. “Today, several tech companies are developing leading closed models. But open source is quickly closing the gap. Last year, Llama 2 was only comparable to an older generation of models behind the frontier. This year, Llama 3 is competitive with the most advanced models and leading in some areas. Starting next year, we expect future Llama models to become the most advanced in the industry. But even before that, Llama is already leading on openness, modifiability, and cost efficiency. Today we’re taking the next steps towards open source AI becoming the industry standard,” he’d said, while announcing the release of the Llama 3.1 405B model.
Zuckerberg had then highlighted how he believed that open-sourcing AI models was good for developers, Meta, and the world at large. Interestingly, back then he’d said that open-source was better for AI safety, which appears diametrically opposed to his stance now. “There is an ongoing debate about the safety of open source AI models, and my view is that open source AI will be safer than the alternatives,” he had then said.
Meta now says that it’ll carefully choose which models it makes open source. It could be because that Meta has realized that AI models are becoming so powerful that it needs to carefully consider which capabilities it puts in the hands of developers. But several open-source models have now been released that are close to the frontier of AI capabilities, including DeepSeek R1 and Kimi, with no obvious ill-effects, so it doesn’t seem plausible that Meta has chosen to change its open-source policy based on that alone.
It’s more likely that Meta’s policies are driven by business considerations. When ChatGPT and Google had launched their AI models, Meta was slightly behind them, and it made sense for Meta to open-source its offerings — it would’ve undercut their pricing, while ensuring greater adoption among the developer community. But over the last year, following the failure of Llama 4, Meta has largely fallen off the map in AI capabilities, and even an open-source Llama model might not find any takers. More recently, Meta has splurged on poaching top researchers from rival labs, and it appears that Meta again wants to build models that compete with the top labs. As such, it would no longer make sense for Meta to open-source these models — it can use its considerable distribution heft to compete against ChatGPT, Google, Anthropic and xAI directly. It remains to be seen which models Meta chooses to open-source, but it appears that the days of Meta open-sourcing its top models could be behind us.