Skip to content

Feature Request: Add logprobs sampling metadata to model database #2743

Description

@wang1970

Feature Request: Add logprobs metadata to model database

Why we need logprobs data

Logprobs are a natural model fingerprint — different models produce distinct token probability distributions on the same prompt, and these distributions are inherently difficult to forge. If an API claims to serve Model A but returns logprobs inconsistent with Model A's baseline, users can immediately detect misrepresentation.

Current problems

  1. No central reference: There's no public dataset of per-model logprobs baselines. Users who suspect mislabeling have nothing to compare against.
  2. High collection cost: Building baselines requires many API calls with consistent prompts — impractical for individual users to do across dozens of models.
  3. models.dev gap: The database already tracks pricing, context limits, and capability flags, but has zero information about logprobs support.

What to add

A logprobs section per model recording:

  • Whether logprobs are available
  • Access method (built-in by default, or requires parameter)

Happy to help collect baselines for a first batch of mainstream models if accepted.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions