Key Takeaways
- Meta and OpenAI’s adoption of open-weight models signals a shift from the dominance of closed-source AI models in the West.
- OpenAI’s planned open-weight model stops short of committing to a full open-source release.
- The success of DeepSeek’s low-cost, high-performance model is pushing Western tech giants to rethink their strategies to maintain AI leadership.
Meta has revealed its new line of open-weight multimodal artificial intelligence models, the Llama 4 herd, which it describes as its most advanced set of models yet.
The move comes as OpenAI “prepares” to release its first open-weight model.
Despite the U.S. tech sector’s long-standing preference for closed-source AI, DeepSeek’s runaway success has driven industry giants to embrace open-weight models.
Meta’s Llama 4
Meta’s new line of AI models furthers CEO Mark Zuckerberg’s vision of making open-source AI “the industry standard.”
On Saturday, April 5, Meta unveiled two multimodal open-weight models, Llama 4 Scout and Maverick, built on a mixture-of-experts (MoE) architecture.
MoE architecture makes large AI models more efficient by dividing the work among specialized “experts” within the model. Instead of having the entire model process every input, only a few selected parts are activated depending on the input.
Llama 4 Scout has 17 billion active parameters and 16 experts designed to fit within a single H100 GPU.
Meanwhile, Llama 4 Maverick is built with a 17 billion active parameter model with 128 experts, designed for more significant use cases and heavy workloads.
Meta claimed Maverick beats OpenAI’s GPT-4o and Alphabet’s Gemini 2.0 on coding, reasoning, multilingual, long-context, and image benchmarks.
It also added that it was competitive with the “much larger DeepSeek V3.1 on coding and reasoning.”
“Our goal is to build the world’s leading AI, open source it, and make it universally accessible so that everyone in the world benefits,” Zuckerberg said in an Instagram reel.
“And I’ve said for a while that I think that open-source AI is going to become the leading models, and with Llama 4, this is starting to happen,” Zuckerberg said.
“Meta AI is getting a big upgrade today,” he added.
OpenAI Stops Short of Open Source
Last week, OpenAI CEO Sam Altman announced the company was preparing to release a “powerful” open-weight model.
The release moves the ChatGPT maker closer to its open research roots but notably avoids committing to full open-source AI.
An open-weight AI model means the developers have released the trained model’s weights—the data that allows the model to make predictions—so anyone can download and use it.
However, to be truly open-source, the code used to train the model, the data it was trained on, and the rights to modify or use it commercially would also need to be released.
With this in mind, Meta’s models are still classed as open-weight but have moved closer to open-source.
OpenAI’s business strategy has dramatically changed since it was founded in 2015, leading to criticism from former founder Elon Musk.
Musk heavily criticized this move, claiming the company had prioritized corporate profits over its original goal of prioritizing AI research for humanity’s welfare.
After leaving the company due to an alleged conflict of interest with his AI work at Tesla, Musk criticized OpenAI’s switch to a capped for-profit model, which allowed it to attract substantial funding from private companies.
DeepSeek’s Influence
Big Tech continues to adapt to the explosive release of R1 from China’s DeepSeek.
DeepSeek’s super-low development costs and open architecture challenge the long-held belief that only tech giants with deep pockets could build frontier models.
The company claimed it trained R1 for just $3 million—a fraction of what competitors like OpenAI and Google have spent on similarly capable systems.
If accurate, that figure sent a clear message to the industry — the barriers to entry for building powerful AI are lower than many assumed.
Since its release, adaptations and customizations of R1 have been embedded into universities, startups, and the public sector.
As China is now home to one of the most competitive open models on the planet, the largest and most valuable AI companies in the U.S. and Europe have been forced to reevaluate how to maintain their global leadership.
Kurt Robson is a London-based reporter at CCN, specialising in the fast-moving worlds of crypto and emerging technology. He began his career covering local news in Cornwall after graduating from Falmouth University with First Class Honours in Journalism. There, he cut his teeth on everything from council meetings to missing swans.
He quickly rose through the ranks to become a frontline journalist at several of the UK’s leading national newspapers. Over the years, he has interviewed musicians and celebrities, reported from courtrooms and crime scenes, and secured multiple front-page exclusives.
Following the upheaval of the COVID-19 pandemic, Kurt shifted his focus to technology journalism—just ahead of the AI boom. With a natural curiosity and a trained eye for emerging trends, he has found a new rhythm in reporting on innovation.
At CCN, Kurt's work focuses on the cutting edge of crypto, blockchain, AI, and the evolving digital world. Drawing on his background in people-first reporting and his deep interest in disruptive tech, Kurt delivers stories that are insightful, entertaining, and human-centric.
