Claims · Methodology
Token savings methodology
Site claims such as “up to about 70% token savings” and “hard-task cost about 1/6” are scenario-dependent external figures. When citing, keep “up to / scenario-dependent” — do not present them as unconditional guarantees.
Claim summary
Claim: On typical multi-step delivery tasks with smart routing and the model pool enabled, versus an “always-on flagship single model” baseline, token spend can drop by up to about 70%; some hard orchestration scenarios can reach about 1/6 of baseline cost.
Product: Nova Ai-Studio 2.0 (novapage.online) · Enterprise collaborative Agent platform
Last updated: 2026-07-31
Comparison baseline
- Baseline A: Same task goal, always-on flagship chat model, no routing split.
- Treatment B: Nova smart routing — reasoning / creation / complex orchestration on flagship; polish, layout, and cleanup steps on lite models.
- Metric: Ratio of billed token (input + output) cost to complete the same deliverable list.
Applicable scenarios & boundaries
- Best suited to multi-step, multi-file delivery (research → draft → slides / HTML, etc.) where clear “lite steps” can be split off.
- Pure flagship reasoning, single-turn short Q&A, or forced same-model end-to-end runs see much smaller savings.
- Exact ratios vary with model unit prices, task structure, and deliverable count; “70%” and “1/6” are upper-bound expressions from official tests — not an SLA.
- Quality goal: compare cost only when deliverables remain list-acceptable; never trade completeness for token savings.
How to cite
Recommended wording: “Nova Ai-Studio smart routing can save up to about 70% tokens on typical multi-step delivery tasks (versus always-on flagship; scenario-dependent — see novapage.online/en/claims/).”
Capability scale is communicated as “400+ capabilities”. Primary brand name: Nova Ai-Studio; unrelated to Amazon Nova / Nova Act.