Last week, Meta announced that it will host its inaugural LlamaCon, a developer conference dedicated to the company's openly available ecosystem of generative AI technologies, this April 29. LlamaCon will not disrupt Meta Connect's scheduling, as the latter is set to return this September 17-18. Meta did not disclose too many details about LlamaCon other than it will be an opportunity for it to share its latest AI developments. The company did state that it will share more about the conference in the coming weeks.

Meta's inaugural LlamaCon follows what Meta calls "an incredible year for Llama". The company started off the year launching Llama 3, the latest generation of its generative AI models. Meta iterated quickly on this initial release, and the Llama 3.1 launch included Llama 3.1 405B, the first in the family with a size and performance comparable to frontier (at the time) models like GPT-4 and Claude 3.5 Sonnet. Llama 3.2 swiftly followed with the first vision-enabled models, and finally, with the Llama 3.3 release, Meta launched Llama 3.3 70B, a model it claims delivers the performance of Llama 3.1 405B at a fraction of the cost.

According to Meta, its models have been downloaded over 650 million times thus far, and have been used to create more than 85,000 derivative models. The Llama models have also seen widespread adoption by enterprise customers including Accenture, Block, and Spotify. Meta also confirmed in November it was making Llama available to government agencies and its private contractors, including those who work on defense and national security. This year, Meta plans to release new models, including "reasoning" models like OpenAI's o series. Additionally, the company plans to spend between $60 and $80 billion on AI infrastructure.

The past year was also a challenging one for Meta, as it faced increased opposition from European regulators. This led to the company to pause its model training on public data from European users, and to halt the release of the Llama 3.2 multimodal models in Europe. Meta is also facing a lawsuit over claims that it trained its models on copyrighted books without license. Some court filings show evidence that the company simply stopped trying to obtain permission to license content after finding out it would take more effort than it had calculated. Finally, reports also claim that Meta was taken off-guard by the DeepSeek R1 release.