AI research laboratory Voltropy has introduced Vast-10M, a new family of frontier AI models designed to process up to 10 million tokens of information in a single context window.
The Irvine, California-based company says Vast-10M represents a major expansion in how much information an artificial intelligence model can analyze at once. According to Voltropy, the model offers approximately ten times the context capacity of many leading frontier AI systems.
Large context windows have become an increasingly important battleground in artificial intelligence as companies seek to develop models that can analyze lengthy documents, extensive datasets, software repositories, and large collections of business information without repeatedly breaking the material into smaller pieces.
“For almost three years, frontier models have been stuck around a million tokens of context,” said Clint Ehrlich, CEO and co-founder of Voltropy.
Ehrlich said Vast-10M could let users give an AI system extremely large collections of information, including something as extensive as the entire U.S. tax code or earnings-call transcripts from every company in the S&P 500.
Voltropy Scalable Attention Powers Vast-10M
At the center of the new AI model family is a technology called Voltropy Scalable Attention, or VSA.
Voltropy describes VSA as a new algorithm that expands the context windows of existing transformer-based AI models while maintaining their reasoning and information-retrieval capabilities.
The technology is being used across three Vast-10M configurations:
Vast-10M-Flash, based on DeepSeek V4 Flash; Vast-10M-Medium, based on GLM 5.2; and Vast-10M-Pro, based on DeepSeek V4 Pro.
Voltropy CTO and co-founder Ted Blackman said previous approaches to extending AI context windows frequently came with a trade-off in model intelligence.
“VSA does the opposite,” Blackman said. “It actually makes models smarter at shorter context lengths.”
According to the company, that improvement lets Vast-10M models compete with leading frontier AI systems even on tasks that don’t require Vast-10M’s full context capacity.
Ten Million Tokens Could Expand Enterprise AI Applications
The ability to analyze substantially more information at once could have implications across industries.
Businesses are increasingly using generative AI to examine financial documents, legal records, research papers, customer information, software code, and internal knowledge bases.
Larger context windows could allow organizations to analyze broader collections of information without relying as heavily on document retrieval systems or repeatedly sending smaller portions of data to an AI model.
Voltropy said Vast-10M-Flash demonstrated competitive performance on the BEAM benchmark, including at the one-million-token level. The company also says the model maintains strong recall performance throughout its expanded context window.
At 10 million tokens, Voltropy reports that Vast-10M-Flash demonstrates greater recall than its underlying DeepSeek model at one million tokens.
The release of Vast-10M comes as AI developers increasingly focus not only on improving model reasoning, but also on increasing the amount of information models can retain and analyze during individual interactions.
Voltropy Expands Its AI Research Efforts
Vast-10M is the first AI model family released by Voltropy PBC, an artificial intelligence research laboratory based in Orange County, California.
The company previously developed LCM: Lossless Context Management, an agentic architecture designed to help AI systems manage extensive amounts of contextual information. Voltropy says it integrated the technology with OpenClaw and Hermes Agent after benchmark testing.
With Vast-10M, Voltropy is bringing its large-scale context management work directly into frontier AI models.
Early access registration for Vast-10M opened September 29 through Voltropy’s website. The company has also published a technical report detailing the model’s underlying technology.
As AI models become capable of working with increasingly large datasets, technologies such as VSA could help push artificial intelligence toward applications requiring analysis of entire corporate, financial, legal, or scientific information environments rather than isolated documents.
For entrepreneurs and businesses adopting AI, the next competitive frontier may therefore involve not only how intelligently a model reasons—but how much information it can understand at once.
