[KongchangAI]
Conceptthinking level / reasoning effort

Reasoning Effort

An abstract concept in large model APIs that controls how much internal reasoning a model performs before producing an answer. It determines whether the model expands long chains of thought, performs self-checking, and conducts multiple rounds of backtracking, and is typically discretized into low, medium, and high levels.

Timeline (last 90 days)

Oct 24

The reasoning effort level (effort/reasoning level) applies to every output token, including thinking, text, and tool calls, not just the final text.

Unverified50%
Oct 16

The reasoning effort tiers of different vendors (such as medium) have no unified definition, and cross-model benchmark tables are basically not comparable if they do not indicate the reasoning tier, context length and tool-calling configuration.

Unverified50%

All Facts (2)

Source Articles