I am experiencing and issues with using turkish, 4.1 and other models are sounds robotic for turkish, so I use 3.1 flash lite as a model and camille from minimax in the french settings, i feel it’s the most compatible accent with the Turkish and also well trained in the meanwhile. But the problem is 3.1’s decision mechanism is not working in the same level all the time. Any recommendations for this case?
Hey @aiclinicsolutions here are the key recommendations for your inconsistent decision-making issue with Gemini 3.1 Flash Lite:
-
Lower the temperature to
0.0–0.3for more deterministic, consistent behavior especially if your agent involves decision branches or function calls. -
Enable Structured Output to ensure reliable function calling and prevent malformed/missing arguments.
-
Switch to Conversation Flow if your agent has more than 3–4 conditional paths.
If you have tried all could you please share the relevant call ID(s) and the agent ID? Also, could you let us know which decision-making mechanism or behaviour you believe is not working as expected?
This information will help us investigate the issue more effectively.
Thank you!