Feature Request: Model Introspection API

API Feedback
1

Hi,

We’re building tools where users bring their own API key and select models by name at runtime.
Before making a responses or chat.completions call, we need to know what a model is and how it can be used.

Today, there is no programmatic way to retrieve this information.

We’re requesting a Model Descriptor (Introspection) API, where providing only a model name returns a machine-readable contract describing that model.

Required fields (blocking issue for BYO-key tools)

{
  "limits": {
    "max_context_tokens": 400000,
    "max_input_tokens": 400000,
    "max_output_tokens": 128000,
    "default_output_tokens": 4096
  }
}

Knowing max_output_tokens is not optional for BYO-key tools.
Correct chunking must be done before the request; relying on failed calls or parsing error messages is brittle and costly.

Parameter contract

"parameters": {
  "temperature": { "min": 0.0, "max": 2.0, "default": 1.0 },
  "top_p": { "min": 0.0, "max": 1.0, "default": 1.0 }
}

This allows clients to automatically adapt when new parameters or ranges are introduced.

Capabilities

"capabilities": {
  "reasoning": true,
  "tools": true,
  "json_schema": true,
  "vision": false
}

Pricing metadata (optional but extremely useful)

"pricing": {
  "input_tokens_per_usd": 1000000,
  "output_tokens_per_usd": 250000
}

Optional metadata

  • description

  • release notes / what’s new

  • deprecated parameters

Why this matters

  • Static model → limit maps are fragile.

  • Parsing error messages to infer limits is unreliable.

  • “Use dated checkpoints” does not work for BYO-key tools where users control the model string.

  • Gemini already expose token limits via API.

This is not a documentation request; it’s a missing platform primitive.

Is a model descriptor / introspection endpoint on the roadmap?

Thanks.

2

Having a programmatic way to retrieve model capabilities would make building BYO-key tools much more robust. Hardcoding token limits or parsing errors is fragile, especially when users can select different models at runtime. A Model Introspection API seems like a logical platform primitive to support dynamic client behavior.

2 Likes