Spanish Firm Launches Quasar 438B, Top European AI Model

Multiverse Computing, a Spanish AI company, has released the Quasar 438B model. It scored 43 on Artificial Analysis's Intelligence Index v4.1.1, ranking highest among European models. This index combines nine tests covering agents, coding, and reasoning. For comparison, Mistral Medium 3.5 scored 30, and NVIDIA Nemotron 3 Ultra scored 38.

Quasar 438B has 438 billion parameters and is designed for enterprise agents, software development, and complex tasks. It supports English and Spanish and is available via the CompactifAI API. Reports suggest it may be based on Zhipu's GLM5.2 model. It's a proprietary model with no open weights, handling text only. Pricing is $0.60 per million input tokens and $1.80 per million output tokens.

For speed, it generates 500 tokens in 15.3 seconds, with an output rate of 176.2 tokens per second and a first-token delay of 1.08 seconds. It scored 75.0 on long-context reasoning tests, with a 1 million token context window, suitable for contracts and technical documents. On Terminal-Bench v2.1, it scored 69.3, handling terminal-based agent tasks.