Meta has officially switched from using its own Llama models to Claude Sonnet for internal coding, signaling a shift in its AI strategy. Meta Drops Llama, Adopts Claude for Coding Meta engineers have confirmed that the company has replaced Llama with Claude Sonnet for internal development tasks. This move comes after Llama 4 faced criticism since its launch, including negative reviews within the first 36 hours and controversies over model rankings. The decision raises questions about Llama 5’s future. Mark Zuckerberg recently established Meta SuperIntelligence Labs (MSL) to develop next-generation AI models alongside the existing GenAI team. However, Meta has not yet clarified whether future models will retain the Llama branding. Llama 4.1 and 4.2 in Development Zuckerberg’s MSL team is nearly complete and aims to deliver a next-gen model within a year. While the exact name remains uncertain, internal documents suggest that Llama 4.1 and 4.2 are already in development. However, a departing Meta employee criticized the company’s cultural challenges, stating that fear-driven management, lack of vision, and toxic competition hinder AI projects like Llama. Despite massive investments and top talent, these issues persist. Claude’s Superior Coding Performance Meta’s decision to switch to Claude Sonnet reflects its acknowledgment of Llama’s shortcomings. Internal feedback confirms improved coding efficiency after the transition, aligning with Claude’s reputation for strong programming capabilities. Claude has gained recognition for its code-generation prowess, with developers praising its performance. For instance, a developer used Claude to write nearly an entire macOS application (Context), contributing less than 1,000 lines of code out of 20,000. For instance, a developer used Claude to write nearly an entire macOS application (Context), contributing less than 1,000 lines of code out of 20,000. For instance, a developer used Claude to write nearly an entire macOS application (Context), contributing less than 1,000 lines of code out of 20,000. Anthropic’s Rapid Growth Anthropic, the company behind Claude, has seen $40 billion in annualized revenue—a nearly fourfold increase since early 2025. This growth is partly driven by Claude Code, launched in February 2025, which has become a key tool for developers. Despite this success, Claude’s user base remains just 1/25th of OpenAI’s, suggesting that a small group of high-impact developers is driving its adoption. FAIR vs. GenAI & MSL Meta’s Fundamental AI Research (FAIR) team, led by Yann LeCun, operates independently from GenAI and MSL. FAIR focuses on open, curiosity-driven research using public data, rather than developing product-level models. Jhu Zeyuan, a FAIR researcher, emphasized that FAIR’s work is not influenced by the success or challenges of GenAI/MSL. Conclusion Meta’s shift to Claude Sonnet highlights the competitive nature of AI development. While Llama’s future remains uncertain, Claude’s rise demonstrates its strong coding capabilities and growing influence in the AI landscape. As Meta continues to invest in next-gen models, the tech community will be watching closely.
