Custom prompt classifiers and reasoning token counts

Custom prompt classifiers gain a quality report, prompt testing, and a Management API

Build, check, and attach a classifier from the dashboard or through /v1.

• Build status, with builds running in the background so saving a classifier returns right away
• Suggested examples drawn from your organization's own request logs
• Scoring that matches intelligent routing, including weighted graded labels
• Classifiers attach to routing policies through the classifier Management API

Read more in the Gateway custom classifiers docs.

Reasoning token counts appear on every SDK-compatible endpoint

OpenAI reasoning models return reasoning summaries by default, and Claude reasoning tokens are derived from billed output. When a request sets no reasoning level, the effort router picks one. Read more in the Gateway reasoning docs.

/v1/responses accepts a plain string input, matching OpenAI's Responses API

Send input as a single string or as untyped message items. Requests written for OpenAI's Responses API in either form are accepted. Read more in the Gateway API reference docs.

BYOK credentials take per-model pricing for cost-ranked routing, and cost attribution

Set a price for each model on a bring-your-own-key credential, and cost-ranked routing ranks those models at your own rates. We also use these rates for better cost attribution and usage analytics. Read more in the Gateway bring your own keys docs.

‍