Meta is accelerating development of its next-generation AI model, Llama 4.X, with plans to release it by the end of 2025, marking one of the first major projects from the newly established Meta Superintelligence Labs (MSL). According to sources familiar with the matter, a dedicated team within TBD—one of four groups comprising MSL—is working to make the models production-ready for the targeted year-end launch. The model is also referred to internally as Llama 4.5 by some team members.
This push comes after Meta’s Llama 4 release in April, which included variants Scout and Maverick, received lukewarm reception from developers who criticized its performance in critical real-world applications such as coding, reasoning, and instruction-following. The TBD team is now tasked with not only developing Llama 4.X but also fixing bugs and reviving the underperforming Llama 4 models.
Meta was also developing an AI model called Behemoth as part of the Llama 4 family, but The Wall Street Journal reported in May that the company postponed its rollout. While Meta declined to provide specific comments, CEO Mark Zuckerberg addressed MSL’s progress during the company’s second-quarter earnings call in July, stating that Meta is “making good progress towards Llama 4.1 and 4.2” while working on next-generation models to “push the frontier in the next year or so.”
Meta Superintelligence Labs was announced by Zuckerberg in an internal memo in June, with the company reorganizing its AI teams by August around four key pillars: training, research, product, and infrastructure. MSL head Alexandr Wang explained in a follow-up memo that the TBD subgroup would focus on “training and scaling large models to achieve superintelligence,” including developing an mysterious “omni model” with no further details provided.
The formation of MSL followed Zuckerberg’s aggressive AI talent acquisition strategy, with reports indicating the company offered multimillion-dollar compensation packages to top researchers from competitors like OpenAI and Google DeepMind. However, the lab is already experiencing turnover—at least eight employees, including researchers, engineers, and a senior product leader, have departed in the two months since MSL’s launch, raising questions about internal challenges at the ambitious new division.
Key Quotes
We’re making good progress towards Llama 4.1 and 4.2, and in parallel, we’re also working on our next generation of models that will push the frontier in the next year or so
Meta CEO Mark Zuckerberg stated this during the company’s second-quarter earnings call in July, providing one of the few public acknowledgments of the company’s AI development timeline and ambitions for advancing the state-of-the-art in AI models.
training and scaling large models to achieve superintelligence
MSL head Alexandr Wang described the mission of the TBD subgroup in an internal memo in August, explicitly framing Meta’s AI ambitions around achieving superintelligence—a level of AI capability that would surpass human intelligence across all domains.
Our Take
Meta’s rushed timeline to ship Llama 4.X by year-end reveals the intense competitive pressure in the AI industry, but also raises concerns about quality versus speed. The fact that teams are simultaneously trying to “fix bugs and revive” the recently released Llama 4 while developing new models suggests potential technical debt and strategic challenges. The early exodus of talent from MSL, despite premium compensation, is particularly telling—it may indicate disagreements over technical direction, unrealistic timelines, or cultural friction within the newly formed organization. The cryptic reference to an “omni model” suggests Meta is pursuing multimodal AI capabilities similar to OpenAI’s GPT-4o, but the lack of details and the postponement of Behemoth indicate possible struggles in execution. Meta’s open-source strategy with Llama has been a differentiator, but if the models continue to underperform in practical applications, developers may increasingly turn to proprietary alternatives, undermining Meta’s AI ecosystem strategy.
Why This Matters
Meta’s accelerated timeline for Llama 4.X represents a critical moment in the AI arms race among tech giants. The lukewarm reception of Llama 4 puts pressure on Meta to deliver competitive models that can match or exceed capabilities from rivals like OpenAI’s GPT-4 and Google’s Gemini. The formation of Meta Superintelligence Labs signals the company’s commitment to achieving artificial general intelligence (AGI) or superintelligence, a goal that could reshape technology and society.
The early staff departures from MSL, despite lucrative compensation packages, suggest potential internal challenges in Meta’s AI strategy—whether related to technical direction, organizational culture, or competitive pressures. For businesses and developers relying on open-source AI models, Meta’s success or failure with Llama 4.X will significantly impact the open-source AI ecosystem. The mysterious “omni model” hints at Meta’s ambitions for multimodal AI capabilities that could integrate text, vision, audio, and other modalities. As Meta races against time to restore confidence in its AI offerings, the stakes extend beyond corporate competition to influence the broader trajectory of AI development and accessibility.
Recommended Reading
For those interested in learning more about artificial intelligence, machine learning, and effective AI communication, here are some excellent resources:
Recommended Reading
Related Stories
- OpenAI’s $157B Valuation: Can It Win the Brutal AI Race?
- Google’s Gemini: A Potential Game-Changer in the AI Race
- Mistral AI Launches Le Chat Assistant for Consumers and Enterprise
- The Artificial Intelligence Race: Rivalry Bathing the World in Data
- Sam Altman’s Bold AI Predictions: AGI, Jobs, and the Future by 2025
Source: https://www.businessinsider.com/meta-superintelligence-lab-llama-4-new-model-launch-year-end-2025-8