---
name: codex-use-gpt-5-6
description: Use when the user asks how to configure Codex to use GPT-5.6, quickly set the default model to gpt-5.6-sol with model_reasoning_effort medium, switch among GPT-5.6 models such as gpt-5.6-sol, gpt-5.6-terra, or gpt-5.6-luna, change reasoning effort, compare max vs ultra reasoning, or explain persistent config versus /model temporary switching.
---

# Codex Use GPT-5.6

Use this skill for Codex's own model configuration, usually in `~/.codex/config.toml`. The default recommendation is `gpt-5.6-sol` with `model_reasoning_effort = "medium"`. Do not use it for application code model constants unless the user explicitly asks for repo code changes.

## Quick Share

To share this skill, give someone the `codex-use-gpt-5-6` folder containing this `SKILL.md`. They should place it at:

```text
~/.codex/skills/codex-use-gpt-5-6/SKILL.md
```

Then they can ask Codex:

```text
Use $codex-use-gpt-5-6 to configure my default model.
```

After installing a new skill, restart or refresh Codex App if `$codex-use-gpt-5-6` is not recognized.

## After Install

On first use after installation, list the available GPT-5.6 models and reasoning efforts before editing config. Then ask whether to use the default recommendation or a different option.

First-use message in Chinese:

```text
可选模型：gpt-5.6-sol、gpt-5.6-terra、gpt-5.6-luna。
可选推理强度：low、medium、high、xhigh、max、ultra。
默认推荐：gpt-5.6-sol + medium。
```

First-use message in English:

```text
Available models: gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna.
Available reasoning efforts: low, medium, high, xhigh, max, ultra.
Default recommendation: gpt-5.6-sol + medium.
```

Tell the user these practical prompts:

```text
Use $codex-use-gpt-5-6 to configure Codex to use GPT-5.6 Sol with medium reasoning.
```

```text
Use $codex-use-gpt-5-6 to switch my Codex default model to gpt-5.6-sol and set model_reasoning_effort to medium.
```

For stronger reasoning:

```text
Use $codex-use-gpt-5-6 to switch Codex to gpt-5.6-sol with max reasoning.
```

```text
Use $codex-use-gpt-5-6 to switch Codex to gpt-5.6-sol with ultra reasoning.
```

Expected default config change:

```toml
model = "gpt-5.6-sol"
model_reasoning_effort = "medium"
```

Tell the user to restart or refresh Codex App after installing the skill if it is not recognized. After the persistent config is updated, tell them to start a new Codex task, and restart/refresh Codex App if the new task still does not use the expected model.

## Activation

Persistent config usually applies to new Codex tasks or after restarting/refreshing Codex App. It may not change the model already running in the current conversation.

If the current Codex surface supports `/model` or a model picker, use that for a current-task switch. If `/model` is sent as a normal chat message or no picker is available, ask the user to start a new task after editing `config.toml`.

General user-facing note in Chinese:

```text
我已经更新了持久配置。这个配置通常会在新会话/新任务中生效；当前会话是否能立即切换取决于 Codex App 是否支持热切模型。如果当前输入 `/model` 没有弹出模型选择器，建议新开一个会话。如果新会话仍未使用新模型，请重启或刷新 Codex App。
```

General user-facing note in English:

```text
I updated the persistent config. This usually takes effect for new Codex conversations/tasks; whether the current conversation can switch immediately depends on whether Codex App supports hot model switching. If `/model` does not open a model picker and is sent as a normal chat message, start a new conversation. If the new conversation still does not use the new model, restart or refresh Codex App.
```

## Config Location

Default config path:

```text
~/.codex/config.toml
```

Useful fields:

```toml
model = "gpt-5.6-sol"
model_reasoning_effort = "medium"
service_tier = "default"
```

Change `review_model` only when the user explicitly asks to change review defaults too.

## GPT-5.6 Models

Prefer checking `~/.codex/model-catalog.json` when available. Known GPT-5.6 model slugs in this environment:

- `gpt-5.6-sol`
- `gpt-5.6-terra`
- `gpt-5.6-luna`

Known `model_reasoning_effort` values:

- `low`
- `medium`
- `high`
- `xhigh`
- `max`
- `ultra`

Use `ultra`, not `ultracode`, unless the local model catalog explicitly says otherwise. In this environment, `gpt-5.6-sol` and `gpt-5.6-terra` support `ultra`; `gpt-5.6-luna` supports up to `max`.

## Common Changes

Default recommendation:

```toml
model = "gpt-5.6-sol"
model_reasoning_effort = "medium"
```

For GPT-5.6 Sol with max reasoning:

```toml
model = "gpt-5.6-sol"
model_reasoning_effort = "max"
```

For GPT-5.6 Sol with ultra reasoning:

```toml
model = "gpt-5.6-sol"
model_reasoning_effort = "ultra"
```

For faster priority service when supported:

```toml
service_tier = "priority"
```

After changing persistent config, mention the activation note above. `/model` is usually a temporary per-task switch; `config.toml` is the persistent default.