@rakyll is working on improving the Gemini API story following:
And
I wanted to share some of my challenges here and see if the community have any others to share:
- The need for
alt=sse
instreamGenerateContent
is very surprising, only documented as a side effect in the Shell example - No actual response body examples are presented for the shell examples
- Tools are documented on separate pages, the first click through provides schema but does not provide examples (Caching | Gemini API | Google AI for Developers which is linked vs Intro to function calling with the Gemini API | Google AI for Developers which is also linked)
REST examples are missing in all sorts of spots:
- Explore audio capabilities with the Gemini API | Google AI for Developers
- Code execution | Gemini API | Google AI for Developers
- Strukturierte Ausgabe mit der Gemini API generieren | Google AI for Developers
etc…
It would be nice if the same examples text generation had carried through consistently. יצירת טקסט באמצעות Gemini API | Google AI for Developers
-
You have to do tons of digging to find model list and figure out which model you should use.
gemini-1.5-pro-exp-0827
vsgemini-1.5-pro-exp-0801
vsgemini-1.5-pro-001
vsgemini-1.5-pro-latest
-
Tool calls changing rules was somewhat confusing in experimental versions, rule changes should be documented: Gemini Pro Experimental 0801 is refusing to run functions - #11 by Sam_Saffron
-
The API overall is quite verbose and nested, it makes it harder to implement compared to anthropic / open ai
Anyone else got feedback? Other feedback I saw here is:
- [Feedback] Focus on Developers First
- Safety settings are a surprise, especially compared to other vendors