# Implicit caching is not working as expected

**URL:** <https://discuss.ai.google.dev/t/implicit-caching-is-not-working-as-expected/111925>\
**Category:** Gemini API\
**Tags:** context\_caching\
**Created:** [December 12, 2025, 4:30pm UTC](https://discuss.ai.google.dev/t/implicit-caching-is-not-working-as-expected/111925 "2025-12-12T16:30:36Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Alireza\_Haghi](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/alireza_haghi/32/31536_2.png) [@Alireza\_Haghi](https://discuss.ai.google.dev/u/Alireza_Haghi)\
**Post date:** [December 12, 2025, 4:30pm UTC](https://discuss.ai.google.dev/t/implicit-caching-is-not-working-as-expected/111925/1 "2025-12-12T16:30:36Z")

</div>

**Context:** I ran a batch job consisting of **100 requests** for text extraction.

**Setup Details:**

- **System Prompt:** Identical and very long across all 100 requests.

- **Output Schema:** Identical and large across all 100 requests.

- **User Input:** The only variable input (different texts for processing).

**Expected Behavior:** Due to the identical, complex prompt and schema, I anticipated a high cache hit rate (significantly above 75%) to minimize computational costs on the fixed components.

**Observed Result:**

- **Total Requests:** 100

- **Cache Hits:** Only **25** (25% utilization).

This low cache utilization is unexpected and inefficient, suggesting the caching mechanism is not optimally recognizing the repeated work (Prompt/Schema overhead).

**Request for Clarification:**

1. What are the **exact caching rules** when the System Prompt and Schema are constant, but the User Input is variable?

2. Why did only 25 requests hit the cache when the costly, non-variable components were repeated 100 times?

[another topic](https://discuss.ai.google.dev/t/system-instruction-and-implicit-caching-question/108831)

 ![image](https://d3qe71uytubmmx.cloudfront.net/original/3X/6/7/672a44f5a5bf7a839469a0b73b5b397940b957f9.png)