The start of ALIA, the Spanish AI model, has been erratic and disappointing. Now we know why

The Spanish AI model ALIA has been criticized for its underwhelming launch, revealing significant challenges in its development. This project, coordinated by the Barcelona Supercomputing Center (BSC-CNS), faced limitations from the outset, leading to inappropriate comparisons with advanced models like ChatGPT. The early release of ALIA-40b has been described as premature, highlighting the complexities and constraints involved in developing large-scale AI models. ALIA-40b is a foundational AI model, trained on a massive and diverse dataset to serve as a base for multiple applications. Despite leveraging the MareNostrum 5 supercomputer, the project encountered limitations. The model's development is more modest compared to proprietary models from the US and emerging open-source models from China. The implications of ALIA's initial shortcomings extend to Spain's ambitions in the AI landscape. The project's goals and development scope are considerably different from those of competitors such as GPT-5 or Gemini 3. Public perception and expectations, shaped by the initial announcement by the Spanish government, clashed with the reality of its capabilities. The contrast between the initial promises and the current capabilities of ALIA underscore the challenges of developing sovereign AI solutions. The project's evolution and future prospects will be critical for Spain's digital sovereignty ambitions in the face of global AI competition.