Updated

Vast-10M (early access)

Voltropy announced its first model family, Vast-10M, in Flash, Medium, and Pro variants, using its Voltropy Scalable Attention method to extend transformer context. The company is offering early access; general availability is not confirmed.

By No Code MBA · Facts checked

Provider
Voltropy
Announcement date
Status at last check
Limited access or preview
Model family
Vast-10M
Capabilities
Text

What changed

Vast-10M is Voltropy's first model family, retrofitting its Voltropy Scalable Attention onto existing models to reach a 10M token native context window. Flash is based on DeepSeek V4.0 Flash, Medium on GLM-5.2 and Pro on DeepSeek V4.0 Pro. Voltropy reports Flash scores 40.20 on BEAM at 10M tokens versus 39.69 for its base model at 1M, and 48.73 at 1M versus 44.03 for Claude Fable 5.1.

Access and availability

Early access to Vast-10M-Flash is available by signing up on Voltropy's early access page. No public API, pricing or general availability date has been announced.

Official access documentation

What you can build

If admitted to early access, test recall on a large document or code corpus against a smaller-context baseline.

Limitations

Voltropy says the models are not more capable than frontier models on the hardest tasks such as frontier mathematics, and calls Vast-10M an intermediate step. Published 10M-token results cover only Flash, not Medium or Pro. No independent benchmarks have been published.

Sources and release history