0 providers0 models

Use-case preset

Email triage and categorization cost calculator

Inbox classification at scale — sales, support, spam, billing.

Email triage classifies inbound messages into categories — sales lead, support ticket, spam, billing, internal — with each call receiving one email (typically 100–400 tokens) and returning a short label or structured JSON output of under 30 tokens. The extreme 95/5 ratio and 2k context reflect the short, self-contained nature of individual emails.

Batch scheduling removes latency pressure entirely. cachedPromptPercent of 50 captures the shared system prompt defining all category labels, example formats, and escalation rules — the largest fixed cost. Because output is a small classification decision, a 7B–14B instruction model is usually sufficient; scaling to 70B adds negligible accuracy on well-defined taxonomies. The dominant cost lever at scale is prompt compression: trimming the category description from 400 to 150 tokens can reduce per-email cost by 20–30%.

Recommended models

8B scale is sufficient for well-defined email classification; lowest cost per email at volume.
Strong instruction following at 8B; reliable structured JSON output for classification labels.
Fast and cheap 7B model; proven on text classification tasks with consistent label adherence.
Compact 9B with good zero-shot classification accuracy on short-text categorization.
Business-oriented pretraining makes it reliable on enterprise email classification taxonomies.