Is Meta Ending Open-Source AI?

In partnership with

Hey there,

One thing is obvious now:
The idea that “AI will stay open” is no longer guaranteed.

Meta is pulling back from open source.

And at the same time, the rest of the industry is trying to build shared standards so things don’t fall apart.

Let’s talk about what’s happening:

Meta Is Stepping Away From Open Source

Meta has been the biggest supporter of open-source AI models.

But now, it’s changing direction.

Reports say Meta has been working on a new AI model called “Avocado”, and unlike Llama, this one will probably be closed-source. Meaning only Meta can fully use or control it.

Why?

Because things inside Meta haven’t been going well:

• Llama 4 has been delayed for months.

• Early versions didn’t impress developers.

• Meta laid off researchers from FAIR.

• Yann LeCun, the strongest open-source voice at Meta, left.

So even though Meta said open source was the future, their actions show the opposite.

And when the biggest open-source supporter steps back, the whole ecosystem feels it.

Other Open Source Models are Still Advancing

While Meta slows down its openness, others are accelerating:

Mistral just released Devstral 2, a 123B model for coding. It’s big, but what stands out is the new Vibe CLI. It’s a tool you install with one command that gives you a coding agent directly in your terminal. 

It can look at your files, understand git status, and help automate tasks through natural language.

DeepSeek took a different route. They released compressed versions of their massive 508B and 345B models using a method called REAP. 

In simple words: They cut the models down by 25–50% while keeping almost the same performance.

That means running very large models becomes cheaper and more accessible.

Even if Meta steps back, others are working to make powerful models cheaper, lighter, and easier to use.

The Industry Is Now Trying to Standardize AI Agents

The Linux Foundation announced the Agentic AI Foundation (AAIF) backed by Anthropic, OpenAI, Block, AWS, Google, Microsoft, and many others.

They contributed three core technologies:

• MCP: a standard for connecting AI models to tools, apps, and data

• Goose: an open-source framework for running AI agents locally

• AGENTS.md: a simple file format that tells coding agents how to operate in a codebase
  

The big AI companies agreed on shared standards so agents work consistently across tools and platforms.

This matters because AI is moving from “chatbots” to agents that perform tasks, coding, automation, file operations, tool usage. Standards prevent the ecosystem from becoming fragmented or unreliable.

It also reduces dependence on any one company’s platform.

How Everything Connects

Here’s the simple pattern:

• Meta is moving away from openness.

• Independent labs are pushing capability and accessibility.

• The rest of the industry is trying to lock down shared standards.

Of course, Meta stepping away from open source doesn’t mean open-source AI is over. But it does signal a shift.

Open-source has never been guaranteed, it’s always depended on whether big companies feel it benefits them.

DeepSeek and Mistral are open today, but plans can change quickly in this industry. As models get bigger, more expensive, and harder to train safely, the pressure to go closed increases.

What You Should Focus On

If you’re building in AI, here’s what matters going forward:

1. Don’t rely on model weights staying open

Meta stepping back is a signal.
Open weights may become the exception, not the norm.

2. Pay attention to agent standards (MCP, AGENTS.md)

These are quickly becoming the foundation for how AI agents work everywhere.

3. Think in terms of workflows, not models

The future isn’t “which model is best.”
It’s “which agent system can reliably complete tasks.”

4. Optimized, compressed models will matter more

DeepSeek’s approach makes frontier-level capability more accessible.
That shifts the competitive landscape.

So what do you think? Is Meta’s shift just one company changing direction, or the first sign that open-source AI won’t stay open forever?

Hit reply and let me know your thoughts.

- AP

Find customers on Roku this holiday season

Now through the end of the year is prime streaming time on Roku, with viewers spending 3.5 hours each day streaming content and shopping online. Roku Ads Manager simplifies campaign setup, lets you segment audiences, and provides real-time reporting. And, you can test creative variants and run shoppable ads to drive purchases directly on-screen.

Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

Use code GET5K now