Gemini.Validation.ThinkingConfig (GeminiEx v0.3.0)

View Source

Validation for thinking configuration parameters based on model capabilities.

Gemini 2.5 series models support thinking budgets with model-specific ranges:

  • 2.5 Pro: 128-32,768 tokens (cannot disable with 0)
  • 2.5 Flash: 0-24,576 tokens (can disable)
  • 2.5 Flash Lite: 0 or 512-24,576 tokens

Special value -1 enables dynamic thinking (model decides budget) for all models.

See: https://ai.google.dev/gemini-api/docs/thinking

Summary

Types

validation_result()

Functions

validate(arg1, model)

Validate complete thinking config including budget and include_thoughts.

validate_budget(budget, model)

Validate thinking budget for a specific model.

Types

validation_result()

@type validation_result() :: :ok | {:error, String.t()}

Functions

validate(arg1, model)

@spec validate(map() | struct(), String.t()) :: validation_result()

Validate complete thinking config including budget and include_thoughts.

Parameters

  • config: Map or ThinkingConfig struct
  • model: Model name string

Returns

  • :ok if valid
  • {:error, message} if invalid

validate_budget(budget, model)

@spec validate_budget(integer(), String.t()) :: validation_result()

Validate thinking budget for a specific model.

Parameters

  • budget: Integer budget value
  • model: Model name string

Returns

  • :ok if valid
  • {:error, message} with helpful error message

Examples

iex> Gemini.Validation.ThinkingConfig.validate_budget(1024, "gemini-2.5-flash")
:ok

iex> Gemini.Validation.ThinkingConfig.validate_budget(0, "gemini-2.5-pro")
{:error, "Gemini 2.5 Pro cannot disable thinking (minimum budget: 128)"}