Multiverse Computing has released HyperNova 60B 2602, a 50% compressed version of OpenAI's gpt-oss-120B, now available for free on Hugging Face. This latest release improves upon the original HyperNova 60B with enhanced tool calling and agentic coding capabilities.
Why Compression Matters
As large language models continue to grow in size and capability, infrastructure requirements become a critical bottleneck. Developers increasingly face constraints when testing, refining, and deploying AI at scale. Multiverse Computing addresses this challenge head-on with their proprietary compression technology: CompactifAI.
The Technology Behind HyperNova
CompactifAI applies quantum-inspired mathematics to analyze and reorganize neural networks, identifying and preserving only their most information-rich components. The results are impressive:
- Up to 95% size reduction possible with only 2-3% precision loss
- Industry-standard compression typically loses 20-30% accuracy with 50-60% reduction
- Models retain advanced reasoning capabilities while requiring a fraction of compute, memory, and power
This approach eliminates the traditional trade-offs between model size, accuracy, and accessibility.
What's New in HyperNova 60B 2602
Building on the original HyperNova 60B released in January 2026, the 2602 variant delivers:
- Improved tool calling capabilities
- Enhanced agentic coding performance
- Same 50% compression ratio as the original
- Maintains frontier-level reasoning despite size reduction
- Completely free and open access
Impact on Developers
For developers working with limited infrastructure or those looking to optimize their AI deployments, HyperNova 60B 2602 offers several key benefits:
- Reduced Hardware Requirements: Run 60B-parameter models on hardware that would typically require 120B capabilities
- Lower Operational Costs: Decreased compute and memory requirements translate to significant cost savings
- Faster Inference: Smaller models process requests more quickly
- Local Deployment Feasibility: Makes running powerful models locally on consumer hardware more realistic
Looking Ahead
Multiverse Computing has announced plans for additional compressed model releases throughout 2026, continuing their commitment to making high-performance AI accessible to the global developer community.
Try It Yourself
HyperNova 60B 2602 is available now on Hugging Face. Whether you're building agentic applications, optimizing existing deployments, or exploring efficient AI solutions, this compressed model provides an opportunity to work with frontier-level reasoning without the infrastructure demands typically required.
This post covers a new model release from February 2026. For more trending AI models, explore our model catalog.