Spanish startup Multiverse Computing has made a bold move in artificial intelligence by unveiling remarkably compact yet powerful AI models, challenging the prevailing momentum toward ever-larger systems. With the debut of the SuperFly and ChickBrain models, the company demonstrates that it's possible to shrink AI model size by up to 95% without any meaningful reduction in performance. Their proprietary compression method, CompactifAI, rooted in quantum physics, is set to redefine what’s possible for edge computing, privacy, and cost-efficiency. This development is backed by notable industry partners and new funding, signaling a major shift in the AI landscape.

The Rise of Compact, High-Performing AI Models

For years, artificial intelligence research has driven a relentless pursuit of larger, more complex models—symbolizing raw technological prowess and dominance in data-rich domains. Multiverse Computing, a Spanish startup with roots in Europe’s quantum research scene, has decisively upended this narrative. On August 14, the company revealed its Model Zoo—a new product suite encompassing two pioneering AI models, SuperFly and ChickBrain. These models, each named after agile animal brains, flaunt extraordinary compression levels achieved through Multiverse’s unique quantum-inspired methodology.

At the core of this breakthrough is the company’s proprietary technology, CompactifAI. By leveraging principles typically found in quantum physics, the team addresses the longstanding tradeoff between model size and performance that has defined AI advancement for years. SuperFly—condensed from Hugging Face’s SmolLM2 135 model—operates using just 94 million parameters, a dramatic reduction from the original's 135 million. Yet, it retains full operational potency.

Engineering Marvel: Quantum-Inspired Compression Technology

Most AI compression strategies to date have drawn heavily on classic computer science, optimizing models via pruning, quantization, or distillation. Multiverse, however, bases its approach on tensor networks—a mathematical structure borrowed from quantum computing, under the hood of CompactifAI. Co-founder Román Orús described this as a “more subtle and refined compression,” achieving up to a 95% reduction in model size without observable performance loss.

The implications are profound. SuperFly is designed for extremely resource-constrained environments and can be seamlessly nestled into smart appliances—from fridges to washing machines—facilitating offline voice and command processing. Imagine instructing your washing machine to "start quick wash" without connecting to the cloud—a leap in both convenience and privacy.

Meanwhile, ChickBrain, with 3.2 billion parameters, derives its architecture from Meta’s Llama 3.1 8B model. This model, engineered for greater dexterity, can run nimbly on consumer-grade laptops such as MacBooks. Notably, it outperforms its “larger” parent model on canonical AI benchmarks—ranging from language comprehension and mathematical reasoning to general knowledge assessments. This achievement dispels the myth that only bulkier neural nets can push accuracy boundaries, redefining performance standards across the AI ecosystem.

Market Embrace and Strategic Partnerships

Multiverse’s breakthrough was not merely an academic feat—it’s been validated by the marketplace. In June 2025, the company closed a €189 million Series B funding round led by Bullhound Capital, with strategic contributions from HP Tech Ventures, Toshiba, and more. Since its founding in 2019, Multiverse has raised close to $250 million, underscoring both confidence in its strategy and the mounting demand for efficient, edge-deployable AI technologies.

Crucially, the firm’s industry collaborations speak volumes. Global technology giants—including Apple, Samsung, Sony, and HP—have all inked partnerships with Multiverse, a rare feat for a startup operating at the bleeding edge of AI and quantum thinking. The reach extends to sectors as diverse as chemicals, automotive, and finance, with clients including BASF, Bosch, and Moody’s. As part of its commitment to accessibility, Multiverse’s compressed models can now be employed by developers via AWS APIs, offering competitive token pricing and straightforward integration into enterprise or consumer applications.

The Economic and Technical Edge of On-Device AI

Shrinking models does more than save storage space. It addresses some of the thorniest issues facing the broader adoption of AI:

Enhanced Privacy: By processing data locally, user information never needs to leave the device, mitigating exposure risks and regulatory burdens.

Reduced Latency: Eliminating the need to constantly ping cloud servers results in faster, more seamless user experiences—critical for mission-critical or real-time applications.

Offline Functionality: Restricted or no internet connectivity no longer hampers smart device intelligence.

Lower Operational Costs: Compression translates directly to reduced compute, storage, and networking expenditures, a bottom-line boon for both manufacturers and enterprise users.

With the proliferation of IoT and smart devices, these features are not just competitive advantages—they could become vital criteria for future product viability. This paradigm may well push the AI industry toward a broader embrace of local, device-level intelligence, breaking the reliance on monolithic cloud data centers.

Strategic Outlook and Investor Takeaways

For forward-thinking investors and technology strategists, Multiverse Computing’s innovation provides a glimpse into the next era of artificial intelligence—one not defined by the pursuit of size, but rather elegance, efficiency, and accessibility. Key competitive takeaways:

Quantum-inspired compression could disrupt the way LLMs are monetized and distributed, giving rise to a new breed of “everywhere” AI, embeddable across the consumer and industrial spectrum.

As partners and clients range from Big Tech to blue-chip industrials, the startup’s market position affords it credibility and reach that transcends the typical boundaries of early-stage AI ventures.

Investors should watch for continued integration of compressed AI into consumer products, as well as strategic moves by incumbents to replicate or license similar compression techniques.

Multiverse’s work represents both a challenge and an invitation: to reimagine AI’s place not as a distant, cloud-bound intelligence but as a seamless, ever-present part of the modern world.