Hey everyone,
Just wanted to share some initial thoughts and raise a few questions regarding the recent updates, specifically the introduction of Gemini 2.5 Flash and LearnLM 2.0.
Gemini 2.5 Flash:
It’s great to see new models being rolled out! However, I noticed that Gemini 2.5 Flash comes with a 500 requests per day limit - vs REQ in day 1500 in Flash 2.0 / Flash 2.0 Thinking.
I understand that with the popularity of Gemini | Google Studio, especially after the impressive Gemini 2.5 Pro release, some adjustments are necessary. Even the Gemini 2.5 Pro limit was reduced from 50 to 25 requests per day, which is understandable given the demand.
However, I couldn’t find any official announcement about Gemini 2.5 Flash in the Changelog. The only information I could find was on the DeepMind website [link to DeepMind 2.5 Flash].
My initial impressions of Gemini 2.5 Flash are a bit mixed. It seems primarily geared towards paid users, likely due to the option to control the “thinking” process. For me, the experience feels like:
-
Thinking ON ~ Gemini 2.0 Flash thinking + Add Grounding with Google Search.
-
Thinking OFF ~ Flash 2.0.
LearnLM 2.0:
Now, about LearnLM 2.0… I’m honestly not even sure if Google is fully aware of this model’s existence! Information about LearnLM 1.5 suggests it’s designed for educators. When hovering over the LearnLM 2.0 option in the interface, there’s no tooltip or additional information provided.
I really hope LearnLM 2.0 comes without usage limits and is freely accessible to educational institutions, as intended with the LearnLM series.
(Maybe no limit | quota in google studio, tell me about this please)
From my limited testing, LearnLM 2.0 feels very similar to Gemini Flash 2.0 – almost like a ctrl+c --> ctrl+v + Tune
model. I haven’t noticed significant differences, except for the context window increase from 32k tokens in LearnLM 1.5 to a reported 1 million tokens in 2.0. I also checked censorship levels, and they seem comparable to Flash 2.0.
Gemini vs. LearnLM for Educational Purposes:
With the latest Gemini versions being quite sophisticated, capable of adopting specialized roles and understanding various domains, including education, I’m starting to question the distinct need for a separate LearnLM model. Gemini itself seems intelligent enough to handle educational tasks effectively.
Seeking LearnLM User Feedback:
Is anyone actively using LearnLM? Are there any real users out there who can share their experiences and insights? I’m genuinely curious to understand its specific purpose and advantages compared to Gemini.
Suggestion for the “Preview” Tab:
Finally, a small suggestion: Could we either rename the “Preview” tab to “LearnLM” to accurately reflect the current offering,
or perhaps start preparing for Gemini 3.0 Preview soon? Let’s just say a peaceful resolution is unlikely if things remain unclear!
More to Come
I’m planning to create a separate topic soon to discuss the limits and quotas in more detail.
Written with Gemini <3.