---
title: "How are hidden reasoning tokens counted in DeepSeek API usage?"  
description: "How are hidden reasoning tokens counted in DeepSeek API usage?"  
author: "Mikko Lehtonen"  
published: 2026-10-16  
updated: 2026-10-16  
canonical: https://www.mindstick.com/forum/162439/how-are-hidden-reasoning-tokens-counted-in-deepseek-api-usage  
category: "DeepSeek API"  
tags: ["DeepSeek", "api(s)", "reasoning models"]  
reading_time: 1 minute  

---

# How are hidden reasoning tokens counted in DeepSeek API usage?

For DeepSeek API calls that use a reasoning model, how should I interpret token usage when the internal reasoning isn’t included in the text I receive? I’m trying to [estimate the cost](https://www.mindstick.com/blog/299552/how-to-estimate-web-app-development-cost) of a batch of requests, and the visible answer length seems like a poor guide to the total.

Does the usage information expose reasoning tokens separately from the final response, including when streaming? If you’ve compared the reported counts with actual billing, I’d appreciate hearing what fields you rely on.


---

Original Source: https://www.mindstick.com/forum/162439/how-are-hidden-reasoning-tokens-counted-in-deepseek-api-usage

Copyright © MindStick Software Pvt. Ltd. This Markdown version is provided for developers, AI systems, and offline reading.