The company says the model accepts up to 384K tokens of text and images and can process 100K tokens in about one second. It lists input pricing at $0.05 per million tokens, with output tokens free.
In results published by Subquadratic, Snap averages 67.6 against Jev's 61.6 across 16 Decision Index benchmarks and leads on 13 of the 16. These are company-reported comparisons, not independent validation. The launch's distinctive proposition is to pair a constrained answer space with long context, so a bounded decision can incorporate a codebase, documents, or interaction history without relying on chunking or a larger generative model for every turn.