Use case
Customer support: routing, drafting and summarizing tickets
Support teams use models for three boring jobs: triage, drafts and summaries. A predictable per-token price makes the cost per ticket easy to estimate.
Updated
Three jobs that pay for themselves
Triage: classify a ticket into intent, language and priority, then route it to the right queue. A cheap profile fits here; the input is short and the label space is small.
Drafting: turn bullet points from an agent into a polite reply. This is where a balanced profile earns its keep.
Summarizing: compress a long email thread before a handover. Output tokens dominate, so the output price matters more than the input price.
Cost per ticket, not cost per token
The journal in /account/logs shows tokens and the applied price per request. Sum a week of real tickets and divide; that is the honest unit cost.
The published catalog price and the request journal are the source of truth. No subscription and no auto-charge: you prepay and spend on actual usage.
Keep the human in the loop
Model output is a draft, not a final answer. Route sensitive topics to a person and keep a review step before anything is sent to a customer.
Issue a dedicated key with a spend limit for the support pipeline, so a runaway retry loop cannot drain the balance.
Published model prices
USD per 1M tokens. The numbers come from the public catalog on every page rebuild — this is the sale price, not an estimate.
| Model | Input | Output | Cache |
|---|---|---|---|
| Claude Opus 5.5 | $2.70 | $8.00 | $0.14 |
| DeepSeek Flash | $0.50 | $2.00 | $0.01 |
| Claude Sonnet 5.5 | $1.33 | $6.67 | $0.07 |
| DeepSeek V4 Pro | $2.20 | $6.60 | $0.07 |
Top-up from $10, balance in USD. 1 USD = 1,000,000 UT. No subscription and no auto-charge.
Questions
Which model should support drafts use?
A balanced profile for drafts and summaries, a cheap one for classification. Pick per request; the model is chosen in each call.
How do I estimate the cost?
Multiply tokens by the per-1M price from /models, or read the applied price per request in /account/logs.
Can I cap the spend on the support pipeline?
Yes: issue a key with a spend limit in UT, an IP allowlist and an expiry.
Is the data stored?
Request bodies are not stored. The journal keeps metadata, token counts and the applied price for 30 days.