Multiverse Computing – a Spain-based deep-tech firm focused on efficient AI and quantum-inspired solutions – saw a week of notable product expansion and strategic positioning. The company’s updates centered on extending its CompactifAI model-compression platform, launching new large language model offerings, and deepening its role in climate-risk analytics.
Multiverse Computing detailed new benchmarks showing its CompactifAI technology enabling a compressed Llama 3.3 70B model to run efficiently on Intel Xeon 6 Performance-core CPUs using vLLM and Intel Advanced Matrix Extensions. The results indicated roughly double throughput versus the baseline model, with latency gains of more than 50% and a disk footprint cut by about half.
Accuracy for the compressed models remained above 97% of the original, with some tasks such as WinoGrande even improving after a targeted post-compression “healing” phase. CompactifAI maintains compatibility with PyTorch and Hugging Face, supporting use cases like retrieval-augmented generation, multimodal reasoning and domain-specific workloads across finance, healthcare and manufacturing.
The company also extended CompactifAI to support a wider set of frontier and mid-size models, including Llama 4 Scout, Llama 3.1 8B, Mistral Small 3.1 and DeepSeek R1, alongside “Slim” variants optimized for efficiency. This breadth, combined with optimized support for Intel Xeon 6 instances across major cloud platforms, positions Multiverse Computing as an infrastructure layer aimed at lowering total cost of ownership for enterprise AI.
In parallel, Multiverse Computing introduced optimized uncensored versions of the GLM 5.1 and Qwen 3.6 27B models, removing weight-level restrictions on sensitive topics while keeping performance on reasoning, coding and general benchmarks largely unchanged. These models are offered via the CompactifAI API and for private deployment, with external safety filters intended to mitigate harmful outputs.
Strategically, the firm announced a commercial partnership with Mitiga Solutions to scale AI-native climate-risk intelligence using Mitiga’s EarthScan platform. Multiverse will serve as technological advisor, applying CompactifAI to deliver efficient, portable analytics on physical climate risks such as floods, wildfires, heat and drought, including asset-level exposure and financial-impact metrics.
The company also highlighted its upcoming participation in the AI4 conference in Las Vegas, where it plans to emphasize “sovereign and efficient AI” and the importance of cost-effective infrastructure in enterprise deployments. Comments from executives underscored a broader shift toward smaller, more efficient models, edge AI and data privacy, alongside Europe’s push for technological sovereignty and long-term quantum computing readiness.
Taken together, the week’s developments reinforced Multiverse Computing’s focus on cost-efficient, CPU-friendly AI infrastructure, expanded its product portfolio into uncensored LLMs and climate-risk intelligence, and strengthened its visibility among industrial and enterprise customers, supporting its position in the competitive AI tools and infrastructure market.