Does native Vapi Gemini support thinkingBudget for...
# supportb
beqodia
06/06/2026, 9:33 AMHi Vapi team,
We're using Gemini through the native Vapi model provider, not through a custom LLM endpoint.
We would like to know whether Vapi currently supports configuring Gemini’s thinking budget for Gemini 2.5 models, especially Gemini 2.5 Flash
Specifically:
1. Is there a way in native Vapi to set:
thinkingConfig.thinkingBudget = 0
2. If yes, where should this be configured? Dashboard, API, advanced model settings, or another field?
3. If native Vapi does not expose this setting, what thinking budget does Vapi use by default for Gemini 2.5 Flash?
4. Is the default Gemini behavior dynamic thinking, provider default, or a Vapi-defined value?
5. Is there any way to confirm in logs whether thinkingBudget is being passed to Google Gemini?
Context:
We are optimizing a real-time voice assistant for low latency. For some use cases, we want to disable thinking or reduce it as much as possible, especially when using Gemini 2.5 Flash in native Vapi.
Expected behavior:
Ability to control or at least know the thinkingBudget used by native Vapi Gemini.
Actual behavior:
We cannot find a visible setting in the Vapi dashboard/API for thinkingBudget.
Thanks.
c
Chiranjeet Mishra
06/06/2026, 9:33 AMHi there,
Thank you for your message. Our team is currently out of the office. We operate Monday through Friday, from 9:00 AM to 8:00 PM Pacific Standard Time (PST).
We’ll get back to you as soon as possible during our normal business hours.
If your message is urgent, please mark it accordingly or include “URGENT” in the subject line, and we’ll do our best to respond promptly.
Warm regards,
Vapi
Customer Support Team
Chiranjeet Mishra
06/06/2026, 9:47 AMHi Beq,
Great question, and you've clearly done your homework on this.
To answer each point directly:
1 and 2. Vapi's native Google provider does not currently expose a thinkingConfig or thinkingBudget field anywhere, whether in the dashboard, API, or advanced model settings. There is no way to set thinkingBudget = 0 through native Vapi today.
3 and 4. Since Vapi sends no thinkingConfig to Google, the model falls back to Google's API default for Gemini 2.5 Flash, which is dynamic thinking. This means the model is actively deciding how many thinking tokens to use on each turn, which will add variable latency to your calls depending on prompt complexity.
Because Vapi doesn't pass thinkingBudget at all, there's nothing thinking-related to confirm in Vapi's logs. You would only see thinking token usage in the raw Gemini API response metadata, which Vapi does not currently surface.
For your latency optimization goal, here are two options worth considering right now:
First, gemini-2.5-flash-lite is already available in Vapi's native model list and has thinking disabled by default. If you haven't tested it, that's the quickest way to remove thinking overhead entirely without changing your setup.
Second, if you specifically need thinkingBudget = 0 on Gemini 2.5 Flash, the only current path is to use a custom LLM endpoint, where you proxy your own Google API calls with that config explicitly set. Vapi supports pointing the model at a custom base URL, so you would retain the rest of your Vapi assistant config.
Native support for thinkingConfig in the Google provider is not available yet. If this is something you'd like to see added, I can flag it as a feature request on your behalf.
Regards,
Chiranjeet
Vapi Support
b
beqodia
06/06/2026, 10:06 AMHi Chiranjeet,
Thank you very much for your detailed response. It clarified several important points and helped us better define the next steps.
As we understand it, Gemini 2.5 Flash-Lite has thinking disabled by default. However, our preference was to use Gemini 2.5 Flash and explicitly disable thinking, because most of our tests and latency evaluations were performed on that model.
Since native Vapi currently does not seem to expose the
thinkingConfigthinkingBudgetthinkingConfig