By skyforbes I was genuinely excited for the Gemini 3 release. With all the panic and hype about how "superior" this model is supposed to be, I expected a massive leap forward.
Here are my main issues:
- It feels… dumber? It’s not that it doesn't know the information. The knowledge base is there. But it feels incredibly lazy regarding structure. It constantly gives half-assed outputs and ignores the structural constraints I set. It feels generic and low-effort compared to the previous goated 2.5.
- Loss of Confidence? 2.5 had a certain confidence and "weight" to its answers. It felt like a powerhouse tool. Gemini 3 lacks that punch. It feels flimsier.
- Fails Miserably as a Study Companion This is my main use case. I need high-level conceptual linking, idea argumentation, and clear theoretical summaries.
- Gemini 2.5: Could take a complex topic, break it down, and link concepts together brilliantly.
- Gemini 3: Skips entire sections, gives surface-level summaries, and fails to link ideas. It honestly feels like a lazy old man trying to get the conversation over with.
- Zero Depth? If I ask 2.5 to explain a complex topic, it breaks it down in a way that actually makes me understand it. It feels tailored to me. Gemini 3 just spits out words. Technically the answer is "correct," but it’s so generic it feels like a glorified Google Search snippet. It lacks the depth and nuance that made 2.5 useful for deep learning.
- Prompts: Almost all my prompts that worked perfectly on 2.5 are now useless. I’m getting consistently bad outputs this didn't happen to any of the previous releases.
I’m confused because everyone is talking about how advanced this is, but I’m just getting stupid outputs. What am I doing wrong exactly? Does Gemini 3 require a completely different prompting style? Or is the model actually regressing in terms of reasoning and following complex instructions?
Any explanation/help would be appreciated 🙂