Hi guys! Trying to pick the best LLM for my assistance, balancing between productivity and low latency, I've understood, that I need to pick GPT 4o (latest) or gpt-4o-mini (and the latest date). But I'd like to ask: What is the best model to use tools? I need only one for now (query tool, knowledge base). When I use 4o-mini, it seems that it doesn't check the knowledge base. Check this call id, messages page (2acc3409-11b8-4283-a898-2cdd11918f4a). There is no tool calls there. When I use 4o latest, I can see this: { "role": "tool_calls", "time": 1749747361028, "message": "", "toolCalls": [ { "id": "call_Lm4A5kF9FLDVi1Zsg7NHmZGX", "type": "function", "function": { "name": "defaultQueryTool", "arguments": "{\n \"knowledgeBaseNames\": [\"Dental_Supply_information\"]\n}" } } ], "secondsFromStart": 20.704 } but there is a problem, that after calling this tool, very often he stoppes and doesn't talk until I'll say something. So, I like more 4o-mini, and I'll use it, but only if he can call tools and read in knowledge base Thanks!!

For reliable Query Tool (knowledge base) usage in Vapi, the recommended model is

gpt-4o

(latest). To ensure tool calls work, configure your assistant like this:

json
"model": {
  "provider": "openai",
  "model": "gpt-4o",
  "toolIds": ["<your-query-tool-id>"]
}

Using

gpt-4o-mini

is not documented to support tool calls or knowledge base access at this time. For best results, stick with

gpt-4o

for tool/knowledge base integration. Sources: - [Using the Query Tool for Knowledge Bases](https://docs.vapi.ai/knowledge-base/using-query-tool) - [Vapi API Reference: Assistants](https://docs.vapi.ai/api-reference/assistants/get)

Sorry, forgot to mention, that maybe there is a problem with retrieving info from knowledge base, because I've just seen a post about "success" posted by @jack-lynq.ai. I have this thing too

Hey! To help track down this issue, could you share: - The call ID - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help us figure out what went wrong!