Foundry Models
A catalog of AI models in Microsoft Foundry that you can discover, compare, and deploy using Azure’s built‑in tools for evaluation, fine‑tuning, and inference
Hello 김교현, Greetings! Thanks for raising this question in the Q&A forum.
Yes, your understanding is correct. The billed_units.search_units value returned in the Cohere Rerank API response is the exact meter that Microsoft Foundry uses for billing, since Foundry passes through Cohere's own usage metering for pay-as-you-go serverless deployments rather than calculating a separate internal count.
Understand what one search unit represents According to Microsoft's own documentation for deploying Cohere Rerank models, a single search unit corresponds to one query evaluated against up to 100 documents. Documents that exceed 500 tokens (including the query length) are split into multiple chunks, and each chunk counts as a separate document toward that 100-document limit. If a single call needs more than 100 documents worth of chunks to be ranked, it consumes multiple search units.
Map this to the API response field The billed_units.search_units field in the Cohere API response reports exactly this count for that specific call, so it is the authoritative usage number for that request. There is no separate token-based or character-based calculation layered on top by Azure for the Rerank models specifically, unlike Cohere's Command or Embed models which bill on tokens.
Verify against your own usage in the portal To confirm this matches what you're actually billed, go to Cost Management + Billing → Cost analysis in the Azure portal, filter by the resource group or resource containing your Cohere-rerank-v4.0-fast deployment, and group by Meter. The consumed quantity shown there for the Rerank meter should track the cumulative search_units values you're seeing returned across your API calls over the same billing period.
If the numbers don't reconcile If you tally up search_units from your calls and it doesn't match the quantity shown in Cost analysis for the same period, that points to a metering discrepancy rather than a misunderstanding of the pricing model, and it's worth opening a support request under Technical with problem type Foundry Models / Cohere so the billing telemetry can be checked against your deployment logs.
If this answer helps you kindly accept the answer which will help others who have similar questions.
Best Regards,
Jerald Felix.