Does EXPLICIT prompt caching work in Gemini 2.5 Pro?

[zerocorebeta]

Issue

Currently, it seems aider uses:

Caching automatically occurs when subsequent requests contain the identical text, images, and cache_control parameter as the first request. All requests must also include the cache_control parameter in the same blocks.

This is how it works in anthropic model.

But in vertex SDK for explicit prompt caching they've this example:

# Create a model instance
model = genai.GenerativeModel.from_cached_content(cached_content=cached_content)

# Now, generate content using the cache.
# The content in the cache will be used as a prefix to your prompt.
response = model.generate_content("Based on the document, what is the main topic?")

print(response.text)

There is no mention if sending the same content without doing genai.GenerativeModel.from_cached_content(cached_content=cached_content) will result in cache being used effectively.

Version and model info

No response