I’m using the Gemini API for financial analysis workflows and I’m seeing weaker results on newer models compared with Gemini 2.5 Pro, so I wanted to ask whether others have seen the same thing and whether there are recommended fixes or migration practices. In one workflow, where I ask for structured company analysis in a single prompt, Gemini 2.5 Pro gives much more thorough and convincing reasoning, while Gemini 3.x produces noticeably shallower justification on the same task. In another workflow, I explicitly instruct the model to choose the conservative figure between weighted-average diluted share count and basic shares outstanding, meaning it should select the larger share count, but Gemini 3.x selected the smaller number and did not justify that choice properly. I’m also seeing problems with long-document coverage: when I provide the full financial-notes section from 10-K or annual report PDFs and ask the model to analyze every note and reference them by number, Gemini 2.5 Pro covers the notes consistently and in order, while Gemini 3.x sometimes skips notes and jumps ahead unpredictably. Are these known issues in Gemini 3.x, and are there specific prompt patterns, API settings, or model configurations that improve reasoning depth, instruction-following, and complete document coverage for this kind of analytical workload? Also, what is the best way to judge when it is safe to migrate from Gemini 2.5 Pro to newer versions for production use in analysis-heavy and document-understanding tasks?
3 Likes
hope to keep 2.5 pro for more time. They haven’t created a model that totally better than 2.5 pro.
1 Like