inadequate-blushI
VAPI•7mo ago
inadequate-blush

Unexpected call analysis model

Hello Vapi team,

Is there any way to configure a model for call analysis for my assistant? (I coudn't find any in the API reference)
Or at least to make sure that the model stays the same as it's important for prompt evaluations?

We have created and tested the call analysis configuration of our assistant against Anthropic’s Claude Sonnet model. But when I actually checked the data of one finished call, I was surprised to see that a totally different model was used:

Call ID: 5e1a7e32-90ca-4976-a44f-01c7d696eab9
    {
      "cost": 0.00053625,
      "type": "analysis",
      "model": {
        "model": "gemini-2.5-flash-preview-04-17",
        "provider": "google"
      },
      "analysisType": "structuredData",
      "promptTokens": 2047,
      "completionTokens": 382
    },


According to the official Vapi documentation, Claude or ChatGPT (as a fallback) must be used for call analysis. Instead a less capable (IMHO) model was chosen for my call.
Was this page helpful?