AI research laboratory Voltropy unveiled Vast-10M, the first frontier AI model with ten million tokens of context. The new model can process ten times as much data at once as the flagship models from Anthropic and OpenAI.
“For almost three years, frontier models have been stuck around a million tokens of context,” said Clint Ehrlich, Voltropy’s CEO and co-founder. “Vast-10M breaks the million token barrier while delivering true frontier intelligence. Users can paste in the entire U.S. tax code or earnings call transcripts for every company in the S&P 500.”
The technology behind Vast-10M is Voltropy Scalable Attention (VSA), a new algorithm which can be used to expand the context windows of existing transformers. VSA delivers ten million tokens of context in three model sizes:
- 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.
“Prior techniques for extending model context degraded intelligence,” said Ted Blackman, Voltropy’s CTO and co-founder. “VSA does the opposite. It actually makes models smarter at shorter context lengths. That boost lets Vast-10M compete with the best models in the world, even on tasks that fit inside those models’ context windows.”
On the industry-standard BEAM benchmark, Vast-10M-Flash beats Anthropic’s flagship model, Fable 5.1, at the one-million-token tier. It achieves parity with OpenAI’s GPT-6 Astra. Vast-10M’s context window is 10x larger than Fable’s and 9.5x larger than Astra’s.
The entirety of Vast-10M’s immense context window is fully functional. On BEAM, Vast-10M-Flash has greater recall ability at ten million tokens than its base DeepSeek model has at one million tokens.
Users can sign up for early access to Vast-10M starting today at www.voltropy.com.
