# How I Choose GPT-5.6 Sol, Terra, Luna, and Effort Levels

> A practical decision guide for choosing a GPT-5.6 model and reasoning effort level based on task difficulty, speed, and iteration cost.

- Author: Kai Wang (AI Kai)
- Published: 2026-08-23
- Updated: 2026-08-23
- Topic: Codex
- Tags: gpt-5-6, codex, model-selection, reasoning-effort
- Canonical: https://hqman.me/blog/gpt56-model-selection/

The GPT-5.6 picker gives me three model families and six effort levels. I use the task's ambiguity, review cost, and expected runtime to choose between them.

![GPT-5.6 model picker](https://hqman.me/media/blog/gpt56-model-selection/model-picker.webp)

## The Three Models

This is my operating heuristic, not an official capability ranking.

**Sol** handles my hardest work: architecture decisions, complex refactors, deep research, multi-step agent tasks, and final code review. I use it when the task is ambiguous or expensive to get wrong.

**Terra** is my daily driver when the plan is clear and the job needs reliable cross-file execution, analysis, automation, or technical writing.

**Luna** is useful for code reading, small fixes, summaries, translations, and batch classification. I reach for it when the task is narrow and easy to verify.

Pricing and quota behavior can change. Check the current product UI or official documentation before estimating cost.

## Effort Levels

Each model offers Low, Medium, High, XHigh, Max, and Ultra.

![GPT-5.6 effort levels](https://hqman.me/media/blog/gpt56-model-selection/effort-levels.webp)

A higher effort level gives the model more reasoning budget and usually takes longer. It does not guarantee a better answer. In my work, I increase effort only after the prompt and task boundary are clear.

## My Default Setup

I use Sol High for most complex coding and multi-file work. For exploration and code reading, I drop to Luna. When a task keeps failing, I move to Sol Max after checking the context and prompt.

I start below Ultra. When Max fails, I inspect the prompt, context, and task split before adding more reasoning budget.