Meta’s Inaugural LlamaCon Reveals the Tech Giant Is Still Trying to Catch Up

If you were hoping, like I was, that Meta’s LlamaCon keynote would unveil the reasoning model it hinted at earlier this month or its teacher model, Behemoth, get ready for disappointment. At its first AI developer conference today, Meta didn’t release any new models. While there were a few updates that help it close the gap in the fast-moving generative AI race, none of them seemed poised to give Meta a real edge.

Every major tech company is in a race to create AI models that can handle complex tasks efficiently, without demanding excessive computing power—and cost. Meta’s strategy has centered on open-source development, giving developers insight into how its models are built and trained. During LlamaCon, Chief Product Officer Chris Cox shared that Llama models have been downloaded 1.2 billion times. Combined with Meta AI integrations across Facebook, Instagram, and WhatsApp, the company remains a significant force in AI—even if it often arrives late or takes a different route.

Here’s a breakdown of today’s Meta announcements and what they mean for its future in AI.

Meta View Rebrands as AI App Before LlamaCon

Meta is rebranding its Meta View smart glasses app into a standalone app for its AI, a move CEO Mark Zuckerberg confirmed on Instagram just hours before the LlamaCon keynote.

The app is available for download now. If searching “Meta AI” doesn’t work—as was the case for me—try looking up “Meta View” instead.

This new app builds on Meta’s chatbot, adding voice interaction features and a “social discovery feed.” Unlike Instagram or Facebook, it doesn’t let you follow friends. Instead, it showcases user-generated content featuring interactions with Meta AI, such as created images, prompts, and responses.

CNBC had hinted at a standalone Meta AI app back in February, but transforming the Meta View app raises broader questions about Meta’s direction in AI and virtual reality. As my colleague and smart glasses expert Scott Stein put it, “Meta making a play for another compelling phone app looks like a way to try to draw more people into the ecosystem faster than making a pitch to get glasses.”

Meta at LlamaCon: No New Llama 4 Models, Behemoth and Reasoning Model Updates Awaited

Meta didn’t unveil the full range of Llama 4 models at LlamaCon; instead, Cox mostly reiterated details we already knew about Scout and Maverick. CNET reached out to Meta for the latest updates on the release of Behemoth and the Llama 4 reasoning model that Zuckerberg introduced earlier this month, but Meta chose not to comment.

Currently, the available models in the Llama 4 family are Scout and Maverick. Scout is a smaller model built to run on a single Nvidia H100 GPU, with a 10-million-token context window, while Maverick offers more power as the next tier up.

There was some confusion when Meta first released the benchmarking scores for Llama 4. The company initially claimed that Maverick outperformed OpenAI’s GPT-4o. Sharp-eyed experts noticed—and the benchmarking organization confirmed—that the tested Maverick model wasn’t the same as the one available; it had been “optimized for conversationality.” Meta denied it had been trained on post-testing data, which could unfairly skew benchmarking results.

Meta’s AI policy states that it does train its models on data shared on Meta Platforms and through content users share with the chatbot. The company also recently removed the opt-out option for European users, so this now applies to them as well. For more details, you can refer to Meta’s full privacy policy.

Meta Previews Llama API for Developer Access to Llama 4

On Tuesday, Meta announced that it will begin previewing its Llama API, a new platform for developers to build Llama applications. Developers can now request early, experimental access to Llama 4 fast inference.

“You’ll be able to take these custom models with you, no locking, ever,” said Manohar Paluri, Meta’s VP of AI. He emphasized that the Llama API will focus on speed, ease of use, and customization. The new Llama 4 models, Scout and Maverick, will be part of the API.

Angela Fan, a research scientist in generative AI at Meta, also pointed out that the API’s privacy policy differs from Meta’s standard AI policy. When using the API, Meta will not train on your inputs (prompts or uploads) or outputs (the generated results), which is beneficial for developers building models for businesses that need to keep data secure.

LlamaCon Helps Meta Catch Up, Lacks Competitive Edge

The announcements at LlamaCon allow Meta to catch up with its competitors, but they don’t give it a competitive edge, which could pose challenges down the line. There’s still no update on when Meta will release Behemoth or the reasoning model it promised with Llama 4.

The Meta View app is a decent effort, but it primarily helps Meta stay on par with other major AI players, such as OpenAI, Claude, and Perplexity, which already have mobile apps. For users of Meta smart glasses, the app’s development could hint at how AI will play a key role in future products.

After the keynote, I felt that Meta is often late to the AI game—OpenAI, Google, and DeepSeek have already released reasoning models. As I mentioned in my review of Meta AI last year, being behind isn’t necessarily a problem if the company makes a strong impact, but so far, that doesn’t seem to be happening.

The most surprising feature was the social discovery feed in the Meta AI app. Given Meta’s expertise in social platforms, the discover/explore page could potentially become a unique (though unlikely) alternative for users to fill their feed with AI content instead of Facebook or Instagram posts. It’s definitely something to keep an eye on as Meta continues to update the app and advance its AI initiatives.

Read the original article on: CNET

Read more: Meta Rolls O ut Limited Teen Accounts on Facebook and Messenger