Benchmarks & research
SubQ Snap is a commercial hosted decision model by Subquadratic that processes up to 384,000 tokens of text and image context for enterprise agent workflows.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
Subquadratic Sparse Attention reduces the computing cost of reading long inputs, allowing the model to evaluate entire codebases or document histories in seconds. For now, access is a public preview constrained by capacity. The vendor supplies all the reported benchmark comparisons against Jev.
The maker shows Snap choosing tools and judging documents from long text and image context. You supply the answer choices and it selects among them. Snap is hosted rather than downloadable, and access to the public preview depends on available capacity.