Find my model

Use case

Ticket routing

Turn a support ticket into a label and a confidence, then let code decide what a person has to see.

What matters

You want a probability, not a paragraph you have to parse. Cost matters because every ticket pays it. Accuracy matters at the threshold you pick, not on a general chatbot leaderboard.

accuracy · structured-output · cost

Top picks

  1. #1Perplexity: Decider Flash V1.2

    Lowest listed input price of the Perplexity deciders, same yes/no, choice, and score questions, and it can see screenshots. Keep it only after it hits your labeled accuracy bar.

  2. #2Perplexity: Decider V1.2

    The larger decider when Flash is not accurate enough, with image input and a 256K context.

  3. #3TypeSafe: Jev 1.14

    Text and JSON tickets that fit in 64K, from the lab that builds only this kind of model.

Budget pick

Perplexity: Decider Flash V1.2

Open-weight pick

Perplexity: Decider Flash V1.2

Pitfalls

  • A chat model that says 'category: billing' will drift. A decision model still needs a threshold you measured.
  • Switching from Decider to Jev changes the probabilities. Re-tune.
  • Perplexity's Orchestrator API is not this. It does not choose the label.

Sample prompt

State: the customer ticket, including any screenshot. Questions: noul 'needs a human', choice of team (billing, access, bug, how-to), score of urgency from 1 to 5. Escalate when p(needs a human) is high or urgency is high.