Quasar 438B (max, based on GLM-5.2)

Quasar 438B is a reasoning-focused large language model developed by Multiverse Computing. Featuring 438 billion parameters, it is derived from the GLM-5.2 base architecture and is specifically optimized for complex, multi-step tasks including code execution, agentic planning, and long-context document research. It was designed to provide a high-performance, European-led alternative for sovereign AI applications, achieving high intelligence scores on indices like the Artificial Analysis Intelligence Index.

The model is constructed using the proprietary CompactifAI compression framework, which reduces the parameter count and computational footprint of the original GLM base while preserving its reasoning depth. This compression allows Quasar 438B to operate with lower latency than traditional models of similar scale, making it particularly suitable for agentic workflows where a single task may require multiple recursive model calls for planning, tool use, and verification.

Key technical capabilities include a 1 million token context window, enabling the stable processing of massive datasets and long-horizon engineering projects. The model provides native support for English and Spanish and features an "always-on" reasoning mode. Users can modulate the depth of pre-response planning through a reasoning effort parameter, with levels reaching up to max for the most intensive logic-heavy requirements. It further supports structured outputs and standard function calling, targeting high-reliability use in software engineering and operational automation.