Meta is racing against time to launch the new generation Llama 4.X AI model before the end of the year
The sense of urgency at Meta stems from the Llama 4 series model released in April this year, which failed to meet market expectations. Reports indicate that developers generally believe Llama 4 performs poorly in practical application scenarios such as coding, reasoning, and instruction following, resulting in a lukewarm market response
Meta is accelerating the development of its next-generation AI model, aiming to launch Llama 4.X by the end of this year.
According to media reports citing two informed sources, a team within Meta's Superintelligence Lab (MSL) named "TBD" is working hard to develop Llama 4.X, with the goal of ensuring that the model reaches a production-ready state by the end of the year.
This sense of urgency stems from the Llama 4 series models released in April this year, which failed to meet market expectations. Reports indicate that developers generally believe Llama 4 performs poorly in practical application scenarios such as coding, reasoning, and instruction following, resulting in a lukewarm market response.
It is reported that part of the current work of the TBD team also includes fixing and improving the existing Llama 4, attempting to address its performance shortcomings. Informed sources also added that Llama 4.X is sometimes referred to internally as Llama 4.5.
In July of this year, the company's CEO Mark Zuckerberg stated during the second-quarter earnings call that the purpose of establishing MSL is to focus on developing next-generation AI models, confirming that "good progress is being made on Llama 4.1 and 4.2," while also working on more cutting-edge next-generation models.
Official Confirmation of Roadmap, Some Models Delayed
MSL was officially established in June this year and completed team restructuring in August.
The restructured AI business operates around four pillars: training, research, products, and infrastructure. MSL head Alexandr Wang explained that the TBD group will be responsible for "training and scaling large models for superintelligence," with tasks that even include developing an "omni model."
Although Meta has not confirmed the specific release date for Llama 4.X, Zuckerberg outlined the company's AI product roadmap in July. He stated at the time that the company is steadily progressing towards Llama 4.1 and 4.2 and is preparing to launch more advanced models in the next year or two.
It is worth noting that the release of Meta's AI models has not always gone smoothly.
Earlier media reports indicated that the company had delayed another AI model named Behemoth, which was originally planned to be launched as part of the Llama 4 series
