I’ve been testing Gemini 2.5 Flash and Pro models in my LangGraph agents, and the experience has been terrible. For example, refining an 8k-character HTML table can take up to 400 seconds. The Flash-Lite model is much faster but produces much worse output.
Has anyone else seen these massive latency spikes with Gemini 2.5 Flash or Pro
Hi @codeonym
Is this issue still occurring? If so, please provide a comparison of the time it took previously versus the time it takes now. Can you also please let us know if this delay is affecting any other tasks.
Also, please test and compare performance with the gemini-3-pro-preview model to see if any improvement is observed.
Thanks
