Apologies for starting yet another thread about AI models in DEVONthink, but I’d be interested in hearing other people’s experiences with the Gemini 3 models in the latest DEVONthink update.
I have previously used Gemini 2.5 Flash via OpenRouter to summarize annotations from PDFs (with a detailed prompt explaining the formatting, detail level, etc.). I was quite impressed by both the speed and the quality of the output. Given everyone’s praise of Gemini 3, I was excited to try whether it would get even better. However, Gemini 3 Flash’s summaries are consistently much shorter and omit important details, even when adjusting my prompt to specifically ask for an extremely detailed summary or requesting a minimum word count.
I’m wondering whether I would get better results through Google’s API, but I don’t really see why that would be the case. In comparison, with the same prompt I still get usable summaries from various OpenAI and Anthropic models, both through their respective APIs and through OpenRouter.
Has anyone made similar experiences with Gemini 3?