Roman Orus

@cgtwts Well, here is @MultiverseQC ‘s CEO with a compressed and uncensored Qwen model running on, hold on, ARDUINO. Who cared about phones now? @elonmusk @sama @MistralAI @AnthropicAI @QualcommVenture

🚀 New: HyperNova 60B 2605 is here. An updated version of our open-source 60B signature model, with stronger coding capabilities and higher accuracy across several benchmarks vs. HyperNova 60B 2602. The headline number: LiveCodeBench 68.68, up from 51.53 in the previous release, and ahead of gpt-oss-120B at 62.75. 📄 Read more here: ow.ly/pmlJ50YVpC2 🔗 Available today on Hugging Face: ow.ly/Cgce50YVpAf #AI #LLM #CompactifAI #HyperNova #MultiverseComputing

📺 @OrusRoman Co-Founder & Chief Scientific Officer of @MultiverseCompu shares the risks of not regulating AI and how policymakers need to closely watch the latest developments of the technology at @rtvenoticias

AI will define our future, but are we ready for its impact? 🤔 Our CSO @OrusRoman , Spain’s representative on the UN AI Scientific Panel 🌍, warns: AI is comparable to the “birth of the atomic bomb” in both power and risk. Yesterday, he joined the UN expert meeting at the Spanish Parliament with PM @sanchezcastejon A new Industrial Revolution is already underway ⚙️🚀 📰 efe.com/ciencia-y-tecn… #AI #ArtificialIntelligence #DeepTech #UN

#AIAMSUMMIT26 | "Cariño, he encogido a la IA”: de la nube al dispositivo industrial Jabier Ruiz (@inetum_es ) · Iraia Ibarzabal (@MultiverseCompu) El diálogo abordó uno de los grandes frenos actuales en la adopción de la IA industrial: pasar del piloto a la implantación real. Modelos sobredimensionados, altos costes, consumo energético y falta de control están ralentizando su despliegue. La conversación puso de relieve la necesidad de una IA más eficiente, estable y predecible, adaptada a casos de uso concretos. No se trata de más IA, sino de una IA que funcione, que sea segura, escalable y capaz de generar valor desde el inicio en entornos industriales reales. #InteligenciaArtificial #Confianza #Innovación #TransformaciónDigital

@MultiverseCompu empresa que aprovecha la algoritmia cuántica para reducir los LLMs de IA. Clave del ecosistema tecnológico apoyado por #SETT, en un contexto donde la cuántica gana peso como vector de innovación, competitividad y autonomía estratégica. sett.gob.es/empresas-parti…

At #MWC26 Barcelona we showed what the future of AI deployment looks like. Our CEO Enrique Lizaso demonstrated a compressed Qwen2-VL model running locally on an @arduino UNO Q 🔌🤖 - no cloud required. The device analyzed a photo taken at the booth in real time 📸⚡ proving that powerful AI can run locally, privately, and efficiently even on ultra-low-power hardware. What radical compression enables: • Multimodal AI on microcontrollers 🧠 • Fully local inference 🔒 • Lower latency & energy use ⚡ • New edge AI applications 🌍 The next generation of AI will be smaller, faster, and more efficient. Learn more 👉 multiversecomputing.com #MWC #EdgeAI #AICompression #CompactifAI #MultiverseComputing

La investigadora @CeliaRogero, la emprendedora #BellaBowring y la empresa @MultiverseQC , premios #EmakumeEkinn 2026 tinyurl.com/y952cpj8

Zorionak a @MultiverseQC, tercer reconocimiento #EmakumeEKINN como 'Empresa que apuesta por el talento y la innovación en femenino'. Lo recoge la CFO Marta García, en nombre de esta empresa líder en software de IA y computación cuántica, de manos de la concejala @aneoyar 🎉

@cgtwts Well, here is @MultiverseQC ‘s CEO with a compressed and uncensored Qwen model running on, hold on, ARDUINO. Who cared about phones now? @elonmusk @sama @MistralAI @AnthropicAI @QualcommVenture

Qwen 3.5 Small models - fully open source - beats models 4x it’s size - 9B model performs on par with GPT OSS 120B while being 13x smaller - outperforms Gemini 3 flash and Claude sonnet 4.5 on select benchmarks - runs on any laptop - even works on a phone - completely free.

📄 Is it possible to reduce an AI model's size by half without sacrificing its intelligence? We explore this challenge in our latest blog post. 👩🏻‍💻 Our Evaluation Engineering Manager, Johanna Angulo, details the Open Source release of HyperNova-60B 2602. We started with a clear conviction: the market doesn't need more massive, general-purpose models, but rather specialized, low-latency solutions that deliver exceptional performance at a fraction of the infrastructure cost. 🚀 The most significant breakthrough in this update is the leap in tool-calling capability. Through our proprietary technology CompactifAI, we've managed to produce a 59B-parameter model capable of running on a single GPU that closely matches the precision of its 117B-parameter parent when interacting with APIs and executing commands. This is the key to making AI agents truly reliable in real-world production environments. You can find all the benchmarks and the full technical analysis at the following link: 👇 ow.ly/8s5Z50YmYvO

un placer como siempre aprender de IA y quantum con @OrusRoman de @MultiverseQC junto a @jemonleon youtu.be/qmUJbUyje8s?si…

Mi enhorabuena a Román Orús por su nombramiento en el Panel Científico Internacional Independiente sobre IA de @UN. Un reconocimiento al talento español y al compromiso con una IA segura, ética e inclusiva.

El español Román Orús (cofundador @MultiverseQC), nombrado por @UN miembro del grupo de 40 expertos del Panel Científico Internacional Independiente sobre IA Descubre el papel de España en el proceso del Panel y en el uso seguro, ético e inclusivo de la IA a escala global 🧵👇

I am Deeply honored to have been appointed to the @UN's Independent International Scientific Panel on AI. AI is reshaping our world, and thoughtful, science-driven dialogue is more important than ever. I am grateful to contribute to this global effort. un.org/independent-in…

🚀 New open-weight release: HyperNova 60B by @MultiverseQC High-performance reasoning, radically lower compute. 🧠 vs Mistral 3 Large: 1.9× better reasoning −92% memory 2.8× throughput +65% faster latency Built for real-world deployment. 🤗 Access the model: huggingface.co/MultiverseComp…

@MultiverseQC trio back at @creativedlab! @SamMugel @BICGipuzkoa

Spoke to @MultiverseQC about how they used quantum physics-inspired methods to compress LLMs and interesting attempts to remove censorship from Chinese open-source models. My latest, with commentary from @thomasqcao : technologyreview.com/2025/11/19/112…

🔬 Big step for AI: @MultiverseQC just hit a breakthrough in compression and de-censorship of DeepSeek R1 Leaner, clearer, more transparent models are coming but so is a major ethics debate. Full control over censorship brings power and risk. 📰 @techreview technologyreview.com/2025/11/19/112…

@QuanticASI That was a peak. What happened to humanity afterwards?

you, me…we're gonna conquer the Second Rome

@QuanticASI Correct 🤌🏻

there won't be a second date but she did learn that time travel is real, dark matter really matters and the universe runs on math