Beginner · Systems

Effort Levels

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. A user-facing control that sets how much compute a model spends reasoning on a request — picking effort instead of picking a model.

Technical Definition

A user-facing control that sets how much compute a model spends reasoning on a request — picking effort instead of picking a model.

How it works

Through 2026, frontier model APIs converged on a simpler dial than the model menu: an effort level — low, medium, high — that controls how long the model thinks before answering. Low effort means near-instant replies cheap enough for autocomplete; high effort means the model reasons at length for hard problems. The shift matters for economics and routing: instead of maintaining a matrix of model tiers, teams route everything to one strong model and tune the effort dial per request, cutting reasoning-token spend dramatically on easy traffic while keeping ceiling performance where it counts. Effort levels also changed product design — progressive answers ('quick reply now, think longer if you ask') became a standard pattern in AI assistants.

Related Concepts

• Local Decision Models — Compact on-device models — like Google's Gemma 4 and Xiaomi's MiMo-V2.6 — that act as traffic controllers, solving tasks locally and calling cloud APIs only for frontier-level reasoning.
• Reasoning Token Economics — The cost discipline of deciding how many hidden reasoning tokens a task deserves — trading answer quality against latency and per-call price.
• Chain-of-Thought (CoT) Prompting — Asking models to show step-by-step reasoning before giving a final answer, improving accuracy on complex tasks.
• Reasoning model — A specialized large language model trained to perform multi-step logical reasoning for complex tasks.