Effort Parameter
A reasoning-depth control parameter exposed at the API level in Claude 3.5 Sonnet, allowing developers to directly adjust the computational resources the model allocates to a given request. It replaces traditional prompt-based guidance for controlling reasoning depth and supports low, medium, and high level settings.
Timeline (last 90 days)
Claude Haiku 5.5 introduces support for an adjustable Effort parameter for the first time, allowing developers to trade off between response quality and cost/speed.
OpenAI's o-series uses the reasoning_effort parameter (low/medium/high) to control inference cost, while Google Gemini uses the thinking_budget token count for control.
All Facts (2)
Claude Haiku 5.5 introduces support for an adjustable Effort parameter for the first time, allowing developers to trade off between response quality and cost/speed.
50%UnverifiedOpenAI's o-series uses the reasoning_effort parameter (low/medium/high) to control inference cost, while Google Gemini uses the thinking_budget token count for control.
50%