Meta Replaces Llama with Claude for Internal AI Development

Meta Platforms Inc. has made a significant shift in its artificial intelligence strategy, with internal teams now using Anthropic's Claude Sonnet instead of its proprietary Llama model for code generation, according to company insiders. This decision follows widespread criticism of Llama4, released earlier this year.

The Fallout from Llama4

Within 36 hours of launch, users reported functionality issues with Llama4, which later became embroiled in controversies surrounding large language model rankings. The negative reception appears to have eroded confidence even within Meta, with engineers reportedly expressing dissatisfaction with the model's performance.

Image source note: The image was generated by AI, and the image licensing service provider is Midjourney

Internal Challenges at Meta AI

A departing employee revealed concerning cultural issues within Meta's Generative AI team, describing an environment of "fear and lack of vision" that has hampered project success. Despite substantial investments in AI talent and infrastructure, these internal challenges continue to impact development efforts.

Meanwhile, CEO Mark Zuckerberg has established the parallel Meta Super Intelligence Lab (MSL), though it remains unclear whether future models will carry the Llama branding. Company memos mention ongoing development of Llama-4.1 and 4.2, suggesting some commitment to the platform continues.

Claude's Rising Popularity

Meta engineers report significant improvements in coding efficiency after switching to Claude Sonnet. The model has gained rapid adoption among developers, with some Apple engineers claiming Claude wrote 95% of code in certain 20,000-line projects.

The shift coincides with remarkable commercial success for Anthropic, which has seen annualized revenue surge to $4 billion - a near fourfold increase since January 2025.

Key Points:

• Meta engineers have transitioned from Llama to Claude for internal code generation
• Llama4 faced severe criticism shortly after release, damaging its reputation
• Cultural issues within Meta's AI teams may be hindering development progress
• Anthropic's revenue has grown dramatically as Claude gains developer adoption
• Future of the Llama brand remains uncertain despite ongoing version development