Thinking Budget
An adjustable parameter in the inference stage of large models that controls the token budget the model devotes to its internal reasoning chain before answering. A higher thinking budget gives the model more reasoning resources, making it suitable for complex tasks such as mathematics and coding, but it may lead to degraded quality in generation-oriented tasks.
Timeline (last 90 days)
Thinking intensity (thinking intensity / reasoning effort) is a tunable parameter exposed to callers by some reasoning models (such as Claude 3.7 Sonnet and the OpenAI o series) to control the depth of the model's internal chain-of-thought reasoning and token consumption.
In multi-agent systems, using lightweight models for sub-agents handling simple tasks and high-reasoning-intensity models for complex reasoning sub-agents is an engineering practice for balancing performance and cost.
All Facts (2)
Thinking intensity (thinking intensity / reasoning effort) is a tunable parameter exposed to callers by some reasoning models (such as Claude 3.7 Sonnet and the OpenAI o series) to control the depth of the model's internal chain-of-thought reasoning and token consumption.
50%UnverifiedIn multi-agent systems, using lightweight models for sub-agents handling simple tasks and high-reasoning-intensity models for complex reasoning sub-agents is an engineering practice for balancing performance and cost.
50%