Hello, I'm having a problem with model billing/routing that I can't seem to resolve, even though my configuration appears to be correct. My configuration (Bot Settings > LLM options): All models are set to Gemini 2.5 Flash-Lite: Default Fast LLM: Gemini 2.5 Flash-Lite Default Best LLM: Gemini 2.5 Flash-Lite Autonomous Language Model: Gemini 2.5 Flash-Lite RAG Language Model: Gemini 2.5 Flash-Lite Fallback LLM: Gemini 2.5 Flash-Lite My Autonomous Node also has “Override default model” and “Override default RAG model” enabled for Gemini 2.5 Flash-Lite. The problem: Despite this, my production logs show the following for every message: Model: Openai:openai • tokens in/out=8537/165 • spend $0.0172 The bot consistently bills via OpenAI, never via Gemini, at a cost of approximately $0.017–$0.02 per message—which corresponds to the rate for a premium OpenAI model, not Flash-Lite. What I’ve already tested: Reposted multiple times after each change Tested in new conversations each time Disabled/re-enabled node overrides: no effect, still OpenAI Disabled “Use Cognitive V2”: the bot stopped responding entirely (108s timeout, 0 tokens charged), so I re-enabled it My questions: Why is the bot billing through OpenAI when none of my settings point to OpenAI? Is this related to Cognitive V2? How can I force the use of Gemini 2.5 Flash-Lite while keeping Cognitive V2 active (since it’s necessary for the bot to function)? I’m also noticing response times that are sometimes very long (up to 58 seconds). Is this related? Thank you in advance for your help.

Heu

Hey ! No, I don't know how to duplicate a bot, and that would mean I'd have to recreate my entire knowledge base, etc. ... https://cdn.discordapp.com/attachments/1519660048668692492/1519724022303031368/message.txt?ex=6a3e98ca&is=6a3d474a&hm=c5689794ab4a1ffb332018c2e1630c8f4f429a972264e41c95fa240894408c29&

No need to create everything from scratch, juste create another bot to see if you face the same problem for now

Based on my investigation, the model you've chosen keeps failing (with code: 503, status: "UNAVAILABLE" "This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later.") which is why the requests are escalated by Cognitive V2 to openAi instead of using your fallback models. I suggest you move from Gemini 2.5 Flash-Lite to Gemini 2.5 Flash and see if that solves your problem. Although it will be slightly more expensive, it's still way cheaper than the fallback model Cognitive V2 currently uses if the first one fails.

OMG thanks you, it's work ! You are the boss 🙌

Happy to help 🤝

hi @wooden-beard-40210 , are you a developer?

Hey 👋 Mostly a bot dev ahah but with codex i built whatever I want as well , I have my agency and I sell AI chatbots, agents and systems