HyperNova 60B 2605 is an open-weight language model developed by Multiverse Computing, a Spanish technology firm specializing in quantum-inspired software. The model is a 50% compressed version of OpenAI's gpt-oss-120B, created using proprietary CompactifAI technology. This compression method utilizes quantum-inspired tensor-network mathematics to reorganize neural networks, significantly reducing the parameter footprint and memory requirements while preserving the reasoning capabilities of the larger base model.
Architecturally, HyperNova 60B 2605 is a Mixture-of-Experts (MoE) model featuring approximately 58.7 billion total parameters, with 4.8 billion parameters active per token during inference. It was designed for high efficiency, capable of running on individual high-end consumer GPUs or dedicated enterprise hardware. The model supports a context window of 131,072 tokens, allowing for the processing of lengthy technical documents and extended agentic workflows.
This release introduces notable improvements in coding performance and general reasoning over previous iterations, such as version 2602. It features native support for tool calling, emitting structured JSON outputs compatible with OpenAI-style function calling schemas. Additionally, the model provides configurable reasoning effort (low, medium, or high), enabling users to adjust the depth of chain-of-thought processing according to the complexity of the task or latency requirements.
HyperNova 60B 2605 is released under the Apache 2.0 license, making the weights available for both academic and commercial use. It is optimized for developer workloads including code generation, agent-augmented tasks, and retrieval-augmented generation (RAG) systems.