Open Source AI is the Path Forward

In the early days of high-performance computing, the major tech companies of the day each invested heavily in developing their own closed source versions of Unix. It was hard to imagine at the time that any other approach could develop such advanced software. Eventually though, open source Linux gained popularity¹ (Editorial note 1)Editorial · 1Known at the timeThe letter opens with history, not with Llama, so that the conclusion feels inherited: closed Unix lost, open Linux won, and AI is next. The analogy is strongest on adoption and weakest on ownership. Linux has no single owner setting its roadmap or licence; Llama has Meta, which trains every release and writes the terms. – initially because it allowed developers to modify its code however they wanted and was more affordable, and over time because it became more advanced, more secure, and had a broader ecosystem supporting more capabilities than any closed Unix. Today, Linux is the industry standard foundation for both cloud computing and the operating systems that run most mobile devices – and we all benefit from superior products because of it.

I believe that AI will develop in a similar way. 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.² (Fact note 2)Fact · 2With hindsightMeta released Llama 4 Scout and Maverick on 5 April 2025. It called Maverick “the best multimodal model in its class”, beating GPT-4o and Gemini 2.0 Flash across what it called a broad range of widely reported benchmarks. The model it set against GPT-4.5, Claude Sonnet 3.7 and Gemini 2.0 Pro, Llama 4 Behemoth, was “still training” and not released. By Meta’s own launch framing, the released models led in their class rather than the industry.SourceMeta AI blogThe Llama 4 herd: The beginning of a new era of natively multimodal AI innovation5 Apr 2025 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. We’re releasing Llama 3.1 405B, the first frontier-level open source AI model³ (Fact note 3)Fact · 3Known at the timeOpen source is Meta’s label for Llama, and the licence sets terms of its own. The Llama 3.1 licence says a licensee whose products had more than 700 million monthly active users on the version’s release date must request a licence from Meta, “which Meta may grant to you in its sole discretion”. The same day, Meta said it had changed the licence to let developers use Llama outputs “to improve other models”.SourcesMeta (llama.com)Llama 3.1 Community License Agreement23 Jul 2024Meta AI blogIntroducing Llama 3.1: Our most capable models to date23 Jul 2024, as well as new and improved Llama 3.1 70B and 8B models. In addition to having significantly better cost/performance relative to closed models, the fact that the 405B model is open will make it the best choice for fine-tuning and distilling smaller models.

[ … ] Not shown: Launch partners, clouds, and why developers want open models.

Meta’s business model is about building the best experiences and services for people. To do this, we must ensure that we always have access to the best technology, and that we’re not locking into a competitor’s closed ecosystem where they can restrict what we build.

One of my formative experiences has been building our services constrained by what Apple will let us build on their platforms.⁴ (Editorial note 4)Editorial · 4Known at the timeMeta’s 2021 founder’s letter complained about living under other platforms’ rules without naming a company; this letter names Apple. It is also the mirror of Steve Jobs’s 2010 case against Flash: Jobs defended a platform owner’s right to decide which layer sits between it and developers, and Zuckerberg writes as the developer who lived under that rule. Open models are insurance against a rival owning the next layer Meta’s products depend on. The cost is that Meta pays to build the layer and lets everyone, rivals included, use it. Between the way they tax developers, the arbitrary rules they apply, and all the product innovations they block from shipping, it’s clear that Meta and many other companies would be freed up to build much better services for people if we could build the best versions of our products and competitors were not able to constrain what we could build. On a philosophical level, this is a major reason why I believe so strongly in building open ecosystems in AI and AR/VR for the next generation of computing.

People often ask if I’m worried about giving up a technical advantage by open sourcing Llama, but I think this misses the big picture for a few reasons:

First, to ensure that we have access to the best technology and aren’t locked into a closed ecosystem over the long term, Llama needs to develop into a full ecosystem of tools, efficiency improvements, silicon optimizations, and other integrations. If we were the only company using Llama, this ecosystem wouldn’t develop and we’d fare no better than the closed variants of Unix.⁵ (Fact note 5)Fact · 5With hindsightMeta’s own download counts rose quickly, though each uses a different basis. On the day of the letter it cited “more than 300 million total downloads of all Llama versions”. On 29 August 2024 it said Llama models were “approaching 350 million downloads”, with “more than 20 million” in the previous month. On 29 April 2025 it said it had “surpassed one billion downloads”.SourcesMeta AI blogIntroducing Llama 3.1: Our most capable models to date23 Jul 2024Meta AI blogWith 10x growth since 2023, Llama is the leading engine of AI innovation29 Aug 2024Meta AI blogEverything we announced at our first-ever LlamaCon29 Apr 2025

Second, I expect AI development will continue to be very competitive, which means that open sourcing any given model isn’t giving away a massive advantage over the next best models at that point in time. The path for Llama to become the industry standard is by being consistently competitive, efficient, and open generation after generation.

Third, a key difference between Meta and closed model providers is that selling access to AI models isn’t our business model⁶ (Editorial note 6)Editorial · 6Known at the timeThis is the classic move of commoditising a complement. Meta earns from ads on its own apps, so cheaper and more widely available models raise the value of what Meta already owns while cutting the margin of firms that sell model access. The argument holds only while models stay something Meta uses rather than something Meta sells.. That means openly releasing Llama doesn’t undercut our revenue, sustainability, or ability to invest in research like it does for closed providers.⁷ (Fact note 7)Fact · 7With hindsightMeta’s spending grew sharply after the letter. In January 2025 it guided to 2025 capital expenditures of “$60-65 billion”. In July 2025 it expected “$66-72 billion”, stated as including principal payments on finance leases, which the January guidance did not mention. It reported $72.22 billion for 2025 on that basis, and guided to “$115-135 billion” for 2026.SourcesMeta, earnings release (Exhibit 99.1)Meta Reports Fourth Quarter and Full Year 2024 Results29 Jan 2025Meta, earnings release (Exhibit 99.1)Meta Reports Second Quarter 2025 Results30 Jul 2025Meta, earnings release (Exhibit 99.1)Meta Reports Fourth Quarter and Full Year 2025 Results28 Jan 2026 (This is one reason several closed providers consistently lobby governments against open source.)

[ … ] Not shown: Meta’s past open source projects, and benefits for the world.

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.⁸ (Fact note 8)Fact · 8With hindsightZuckerberg qualified this a year later. In his “Personal Superintelligence” letter of 30 July 2025 he wrote that superintelligence “will raise novel safety concerns”, and that Meta would “need to be rigorous about mitigating these risks and careful about what we choose to open source”.SourceMetaPersonal Superintelligence30 Jul 2025 I think governments will conclude it’s in their interest to support open source because it will make the world more prosperous and safer.

[ … ] Not shown: Harm, China, and Meta’s new ecosystem approach.

I believe the Llama 3.1 release will be an inflection point in the industry where most developers begin to primarily use open source⁹ (Editorial note 9)Editorial · 9With hindsightDownloads show interest, not which models developers use most, so Meta’s figures cannot settle this. What is clear is that Meta changed its own organisation within a year: on 30 June 2025 Zuckerberg put its AI teams under a new Meta Superintelligence Labs, with a new lab for “the next generation of our models”.SourcesMeta AI blogEverything we announced at our first-ever LlamaCon29 Apr 2025CNBCMark Zuckerberg announces creation of Meta Superintelligence Labs. Read the memo30 Jun 2025, and I expect that approach to only grow from here. I hope you’ll join us on this journey to bring the benefits of AI to everyone in the world.

[ … ] Not shown: Where to download the models, and a sign-off.

MZ