Do Son 
Following the release of the Llama 4 series of AI models this April, Meta’s trajectory in the AI domain has not progressed as smoothly as anticipated. Some developers criticized Llama 4 for underperforming in real-world tasks such as code generation, reasoning, and instruction adherence, falling short when compared to rival offerings. In an effort to regain momentum, Meta has positioned its newly established Meta Superintelligence Labs (MSL) as the driving force behind the accelerated development of a more powerful Llama 4.X, with the goal of unveiling it publicly before the end of the year.
According to insiders, Llama 4.X — reportedly codenamed Llama 4.5 — is being developed by a specialized team within MSL known as TBD. Rather than serving merely as an incremental update, this release is designed to address critical computational flaws and performance limitations in the Llama 4 series, while reinforcing its core capabilities. Through these refinements, Meta hopes to restore its standing within the open-source large language model community.
Originally, Meta envisioned two variants of Llama 4, dubbed “Scout” and “Maverick”, but market reception proved tepid. Another heavyweight model, “Behemoth,” has faced internal delays, underscoring the challenges Meta has encountered in pushing forward with the Llama 4 roadmap.
To reinvigorate its AI strategy, Meta officially launched Meta Superintelligence Labs in June, appointing former Scale AI CEO Alexandr Wang as its head. The lab is structured around four pillars — training, research, product, and infrastructure — with the TBD team specifically tasked with training and scaling large models in pursuit of artificial superintelligence. Internal memos have even referenced the ambition of building an “omni model,” though details remain undisclosed.
During Meta’s Q2 FY2025 earnings call, CEO Mark Zuckerberg emphasized the company’s parallel work on Llama 4.1 and 4.2, while simultaneously preparing for the next generation of AI models, with the ambition of sustaining rapid innovation over the coming year. This aligns with Meta’s recent aggressive recruitment strategy, which includes multimillion-dollar compensation packages to attract researchers from institutions such as Google DeepMind and OpenAI.
Yet, less than two months after its inception, MSL has already faced notable attrition, with reports indicating that at least eight researchers, engineers, and product leads have departed. This reflects both the intense global demand for AI expertise — with constant poaching offers at lucrative terms — and the considerable challenges Meta faces in assembling and retaining a cohesive “superintelligence team.”
Nevertheless, Meta remains determined to showcase its resolve by releasing Llama 4.X before year’s end. If the update can deliver tangible improvements in both performance and stability, it could strengthen Llama’s influence within the open-source community while laying a more solid foundation for Meta’s broader AI ecosystem.
Related Posts:
- Meta Launches Llama API Toolkit for Developers
- Meta’s AI Push: From Llama to Chatbot, Subscription Model Coming
- Legacy vBulletin 4.x Patch Backfires: RCE via Signed Base64 Payloads and a Full PoC
- Meta Eyes PlayAI Acquisition: Bolstering AI Voice Tech & Superintelligence Ambitions
Support Our Threat Intelligence
If you find our CVE report and cybersecurity news helpful, consider supporting our work.
Buy Me a Coffee
PayPal
