bumping this. we still have about 1-3% | ElevenLabs | Page 1

normal carbon Jan 7, 2026, 4:17 PM

#

Hey, please add something like this to the system prompt for now " Never include <thinking>, </thinking>, or any reasoning tags in your responses."

It seems like these conversations are successful , it has failed status because of the evaluation criteria that you set. The <thinking> tags appearing in responses is a known issue with reasoning-enabled LLM models (Claude, Gemini 3, etc.).

You can also switch to Gemini 2.5 Flash and keep thinking budget toggle off. We will work on getting rid of these tags on our end though

fringe sable Jan 7, 2026, 4:18 PM

#

i already did this and it still thinks all the time - this is already a known issue though at least 😅 ?

#

we made the eval criteria because of how much it was happening

#

the more context its handling, the more it happens

normal carbon Jan 7, 2026, 4:19 PM

#

yeap its a known issue , however it shouldn't happen with gemini 2.5 Flash for example

#

if you keep thinking budget off

#bumping this. we still have about 1-3%