We have been experiencing significantly slower response times on the newly released 2.5 pro 0605 model. For comparison, on our testbed with structured output tasks, gemini 0506 has an average response time of 15.4 seconds, while 2.5 pro 0605 exceeds our 180 second timeout limit 28/30 times. We have also tried to reduce this by lowering the thinking budget to 2048 on 0605, but even on this trial, 26/30 trials exceeded our 180 second limit. We are using langchain google genai library with the following library versions:
langchain: 0.3.24
langchain-google-genai: 2.1.4
google-genai: 1.16.1
does anyone face a similar problem? Do you have any suggestions/solutions?
3 Likes
Firstly Welcome to the Google AI Forum! 
Thank you for reaching out and providing detailed information about your experience with the Gemini 2.5 Pro 0605 model.
We understand that you’re encountering frequent timeouts, even after adjusting the thinking budget. This issue has been noted by other users as well, particularly with large prompts or complex tasks. While the 0605 model introduces enhancements in reasoning and coding, it may require more processing time, especially for intricate requests.
To mitigate this:
Optimize Prompt Design: Break down large tasks into smaller, more manageable prompts to reduce processing time.
Monitor API Usage: Keep an eye on your API usage to ensure you’re within the recommended limits and avoid potential throttling.
If the issue persists, please provide additional details about your implementation, such as the specific tasks you’re performing and any error messages received. This information will help us assist you further.
We appreciate your patience and understanding as we work to improve the Gemini 2.5 Pro model.
1 Like
Two days ago I migrated one of my “creative writing assistant” prompts over to Gemini 2.5 Pro Preview from the older 05-06 mdel, and certainly noted a slowdown in response with errors in Google Chrome [This page is slowing down Chrome [Wait] [Stop]] which I had to press [Wait 2-4 times before I got results - after waiting 30-80 seconds. Today however on the 13th, the response is a lot faster and I am getting not errors, the model also seems to be doing better with the task. Just an observation. I hope it continues!
1 Like
We have the same issue.
Recently switched to Gemini 2.5 and experiencing significant response delays
