Multiverse Computing Claims Compressed Model Beats OpenAI Original on Speed and Cost

Multiverse Computing, a Spanish firm specializing in quantum-inspired AI compression, claims its latest model delivers faster responses at lower cost than the OpenAI foundation model it was built upon—a technical milestone that could accelerate enterprise adoption of compressed alternatives to frontier large language models.

The company's HyperNova 60B 2602 is derived from gpt-oss-120b, an OpenAI model whose underlying code is publicly available. Multiverse says the compressed version now outperforms the original on speed and operational expense, advantages the firm positions as particularly relevant for agentic coding workflows where AI autonomously executes complex, multi-step programming tasks.

The announcement arrives as the small-model ecosystem gains momentum. Earlier this week, France's Mistral AI updated its compact model family with Mistral Small 4, optimized simultaneously for general chat, coding, agentic tasks, and reasoning. Mistral also released Forge, a system enabling enterprises to build custom models and select performance tradeoffs tailored to specific use cases.

Multiverse has embedded real-time usage monitoring into its API, reflecting enterprise demand for visibility into compute costs—a central driver behind interest in smaller models as alternatives to resource-intensive LLMs. The firm is positioning edge deployment and reduced infrastructure overhead as strategic advantages over centralized, large-scale architectures.

(Multiverse Computing has not disclosed independent benchmarking or third-party validation of its performance claims. The company operates at the intersection of quantum computing research and classical AI optimization, a niche that remains commercially unproven at scale.)

The competitive landscape for compressed and small models is intensifying as enterprises reassess the cost-performance calculus of frontier LLMs. Mistral's dual release of an updated small model and a customization platform signals a broader industry shift toward modular, use-case-specific AI architectures rather than one-size-fits-all large models. Multiverse's claim to outperform an OpenAI base model—if validated—would represent a technical inflection point, suggesting that compression techniques may now rival or exceed the efficiency of the original architectures they derive from.