Multiverse Computing Launches Quasar 438B Reasoning Model for Enterprise Agents

Multiverse Computing has released Quasar 438B, a large-scale reasoning model engineered for enterprise agents and complex coding tasks. The model is designed to support multi-step workflows requiring planning, tool use, and code execution.

Available through the CompactifAI API, Quasar 438B supports both English and Spanish to serve international enterprises. The model specifically targets organizations deploying technical copilots, software development agents, research systems, and workflow automation. This release marks the first large-scale model produced by Multiverse Computing.

Performance Benchmarks

According to the company, Quasar 438B scored 43 on the Artificial Analysis Intelligence Index v4.1.1, a composite metric based on nine evaluations including SciCode, GPQA Diamond, and Humanity's Last Exam. In this index, the model outperformed NVIDIA Nemotron 3 Ultra, which scored 38, and Mistral Medium 3.5, which scored 30. The company notes that only Gemini 3.7 Flash is both faster and higher-scoring than Quasar on this specific index.

In terms of raw scale, Quasar 438B features 438 billion parameters, which is 112 billion more than Nemotron 3 Ultra. On the AA-LCR long-context reasoning benchmark, the model achieved a score of 75.0. For coding-specific performance, it recorded a 69.3 on Terminal-Bench v2.1.

Operational Efficiency

To ensure the model can be integrated into interactive products rather than being limited to batch processing, Multiverse Computing focused on response latency. The model returns 500 tokens, including thinking time, in 15.3 seconds. This speed is intended to facilitate real-time interaction for technical users and developers.

Industry Implications

The launch of Quasar 438B reflects a broader industry shift toward "reasoning" models that prioritize planning and execution over simple text generation. By targeting the enterprise agent market, Multiverse Computing is positioning itself to compete in the high-end technical automation space, where long-context reasoning and precise code execution are critical for reliability.

Future Outlook

While the model shows strong results in general intelligence and long-context reasoning, the company has identified the Terminal-Bench score as the area with the most headroom for future improvement. Observers will likely watch how the model performs in real-world enterprise deployments compared to the frontier group of reasoning models as it becomes more widely available via the CompactifAI API.