Vast-10M: The First Frontier LLM with 10M Tokens of Context (www.voltropy.com)

Voltropy has launched Vast-10M, a groundbreaking large language model (LLM) that features a remarkable ten-million-token native context window, breaking past the existing one-million-token limit seen in models from OpenAI and Anthropic. This advancement is driven by a novel algorithm called Voltropy Scalable Attention (VSA), which enhances reasoning capabilities even at shorter contexts while vastly increasing recall and analysis abilities in long-context scenarios. In tests, Vast-10M-Flash notably outperformed Fable 5.1 and matched GPT-6 Astra, proving its prowess in processing extensive data sets such as the full U.S. tax code or extensive corporate documents. The significance of Vast-10M extends beyond its impressive context length; it represents a shift in how AI models can effectively learn and adapt without being retrained at inference time. By integrating VSA, these models can potentially acquire new skills and knowledge on-the-fly, positioning them for continuous learning and making them more practical for tasks traditionally dependent on custom retrieval systems. With this innovative architecture, Vast-10M promises not only to enhance current AI functionalities but also to pave the way for safer, more reliable AI systems that allow user control over knowledge retention and degradation.