- Published on
Multiverse Computing Launches Quasar 438B
- Authors
- Name
- DeepHorus
Multiverse Computing has launched Quasar 438B, its flagship bilingual reasoning model designed for enterprise-scale agents and coding. This launch marks the company's first large model release, positioning it as a significant contender in the competitive AI landscape. The model supports English and Spanish, aiming to provide robust capabilities for complex, tool-using applications within businesses.
Quasar 438B is a 438-billion-parameter model that achieved the highest Artificial Analysis Intelligence Index score among European models tested. It scored 43 on the index, outperforming Mistral Medium 3.5, which scored 30, and NVIDIA Nemotron 3 Ultra, which scored 38. This result signifies a major step for European sovereign AI, demonstrating its ability to compete with leading models from the United States and China.
In terms of speed, the model generated 500 output tokens in 15.3 seconds. This performance is faster than Mistral Medium 3.5, which took 18.8 seconds for the same output, while delivering a 13-point higher intelligence score. Such speed is critical for agentic systems where multiple model calls in sequence can compound delays, affecting real-world productivity.
The model's capabilities are particularly strong in long-context reasoning and coding. It scored 75.0 on the Artificial Analysis Long Context Reasoning benchmark, matching Grok 4.6 (high) and coming within one point of Claude Opus 5. Furthermore, on Terminal-Bench v2.1, which tests agents in real terminal environments, Quasar scored 69.3, placing it significantly ahead of both Mistral Medium 3.5 and Nemotron 3 Ultra.
Quasar is available through the CompactifAI API, with planned applications including software engineering, operational automation, and document-heavy research. The launch addresses a key need for European enterprises for models that can handle complex, multi-step tasks while also providing greater regional choice in powerful AI development.
Enrique Lizaso, co-founder and CEO of Multiverse Computing, stated that Quasar shows European AI developers do not have to choose between reasoning performance and speed. He emphasized that the model is built specifically for enterprise agents and coding, bringing together the requirements of complex task execution and tool use.
While Quasar trails the top frontier results on some benchmarks, such as Terminal-Bench where Claude Opus 5 leads, its launch establishes a new high-performance option within the 400-billion-plus parameter class. The development highlights Multiverse Computing's focus on efficient, deployable AI models that balance scale with the low latency required for interactive enterprise products.
