Find my model

Use case

Content moderation

Score a post, comment, or image for a policy and auto-hide only the confident hits.

What matters

False positives annoy people, false negatives create harm. You need a score you can threshold, and a path to a person for the middle.

accuracy · structured-output · vision · cost

Top picks

  1. #1Perplexity: Decider V1.2

    Image plus text, and a yes/no probability so you can hide only the high-confidence cases.

  2. #2Perplexity: Decider Flash V1.2

    The cheaper pass for clear-cut, high-volume queues after you measure precision.

  3. #3TypeSafe: Jev 1.14

    Text-only policies where you want TypeSafe's calibrated score and can stay inside 64K.

Budget pick

Perplexity: Decider Flash V1.2

Open-weight pick

Perplexity: Decider V1.2

Pitfalls

  • Don't hide content on a vendor's benchmark blog post. Measure precision on your policy.
  • Images need a model that lists image input. Jev 1.14 does not.

Sample prompt

State: the post text and image. noul 'violates the hate policy'. score severity 0-3. Auto-hide only above a probability you set from a labeled week of data.