Use case
Bulk translation and localization drafts
Machine translation is a drafting step: fast to produce, cheap to run, and always post-edited by a person for anything customer-facing.
Updated
Batch translation in plain steps
Split documents into stable segments, send each with the target language, and keep the source segment as the key for re-runs.
Ask for the translation only, without added commentary, so the output is machine-checkable.
Cost and repeatability
Translation is input-plus-output tokens. Use the catalog prices to estimate per-document cost, then verify against /account/logs.
Re-running an identical segment with the same request id does not bill twice. Keep request ids stable for idempotent batch jobs.
Where a human is required
Model output is not certified translation. Anything legal, contractual or customer-facing goes through a human reviewer.
Use a spend-limited key for the batch so a malformed loop cannot overspend 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 drafts translations?
A balanced profile handles most product text; a cheap one is enough for short, simple segments. Compare prices on /models.
Is repeat billing possible?
A stable request id for the same segment is not billed twice. A new id means a new request and a new charge.
Can I cap the batch cost?
Yes, with a spend limit on the key before the run.
Are translations machine-only?
Treat them as drafts. Customer-facing and legal text goes through human review.