NO - NO INDIVIDUAL RIG RUNS IT
SubQ Snap has no published quantization that fits a rig one person can buy, so there's no local-hardware recommendation and no break-even to compute. The honest answer is the per-provider API cost above.
The first decision model that reads like a long-context model. SubQ Snap (Subquadratic, October 12, public preview) applies the company’s sparse-attention research to the decision-model lane: a single decision may take up to 384,000 tokens of context, mixing text and images, and a 100,000-token request comes back in under a second. Subquadratic says Snap scores 61.7 on the vendor’s own Decision Index against Jev’s 57.9, which makes the honest reading: a claim from the same team that sells the index. The pricing is the aggressive part. 5 cents per million input tokens with output tokens free, roughly a fifth of the cheapest frontier decision tier, priced for agents that gate every tool call. Output being free is a natural fit for a model that does not really generate text: it picks. Closed for now. No weights, API only, model id subq-snap-preview under a public preview label; evaluate against your own labeled decisions before trusting the index gap.
See the
model card
for the full architecture notes and any cloud subscription plans.