Gemini-1.5-Flash-002 Discontinuation date is September 24, 2025? It's too short. We hopt at least 5 years

Gemini API
1

It shows Gemini-1.5-Flash-002 Discontinuation date is September 24, 2025.
This makes application developers worried about it.
We build up some applications based on this model’s API and we hope it will keep for at least 5 years. Is it the same for both API and vertex?
Hopefully, new model will be better than old model. However, some features or safety check may be diffrent for new models. It’s very important for our applications, please make the API model live at least for 5 years. On the other hand, OpenAI doesn’t not stop supporting for old models.

1 Like
2

How did you find that out? I’m wondering when 1.5-Flash (non 002) will discontinue (if there is a discontinuation date)

3

For various reasons, Google targets the lifetime of these models at once year and announce the planned deprecation date at the time it is launched.

The biggest reason is that, unlike traditional software, it isn’t possible to “fix” bugs in a model.

And certainly at this point in the development of LLMs, expecting it to remain unchanged for 5 years is somewhat unreasonable.

The API itself is unlikely to change. The model is the bigger concern, I agree.

An even bigger concern are embedding models, which cause a major problem when updated since the embeddings aren’t compatible. These tend to have a slightly longer valid date.

Ummm… yes. Yes they do. Sometimes with just 6 months notice.
So much so that they have an entire page about their legacy and deprecated models.

4

Vertex AI publishes this info for each version of each model.

I assume you mean “gemini-1.5-flash”? This is just an alias for the latest stable (ie - numbered) version, which is currently the 002 version. There hasn’t been an announced deprecation date of Gemini 1.5 family as a whole, tho we’re only assuming there will be followon versions to 002.

2 Likes
6

It depends on the application. There may be some bugs for old version, but the old one may be more stable and perform as we expected. We don’t need Google fix the bug, and just leave it as it is. Some content related to sex or polics may be over-correct for the new models even the new one has the better overall performance. This is not what we want.