Meta unveiled a new flagship artificial intelligence model called Muse Spark, the company’s first under the revamped A.I. division spearheaded by Mark Zuckerberg. The model outperformed Meta’s previous A.I. models on writing and reasoning tests, coming close to the top models from competitors like Google and OpenAI, but fell short in coding ability, an area that Anthropic and others are focusing on in the race for A.I. supremacy. This announcement follows Anthropic’s decision to withhold its latest A.I. model, Mythos, due to cybersecurity concerns.
Meta’s efforts to catch up with A.I. leaders like Anthropic, Google, and OpenAI are being closely monitored. Zuckerberg has invested billions in recruiting A.I. specialists to form a “superintelligence” lab, under the leadership of Alexandr Wang. The company plans to invest $600 billion in new data centers to stay ahead in the A.I. race, with a forecasted expenditure of up to $135 billion this year, nearly double that of last year.
Muse Spark, internally referred to as Avocado, is now available on Meta’s A.I. app and will soon be integrated into WhatsApp, Instagram, and A.I. smart glasses. Unlike previous models, Muse Spark’s code is closed for now, with the possibility of open-sourcing parts in the future. Zuckerberg reshaped Meta’s A.I. division after the underwhelming release of its last A.I. model, Llama 4, leading to the departure of past executives and the recruitment of new talent from companies like OpenAI.
Meta aims to create a “superintelligent” form of A.I., a groundbreaking technology envisioned as the ultimate personal companion. The development of Muse Spark faced internal challenges and delays, with Zuckerberg managing expectations for the model’s capabilities. Meta is still working on improving coding skills and has plans for larger, more powerful models in the pipeline, with the next model codenamed Watermelon already in development.
