Which model should I choose?
A model is the “brain” an agent uses to think about your task. The same agent (Claude Code, Codex and others) can run on different models: some are smarter, others are faster. In the pane settings window SHIPIT shows models under their real names. This page explains what those names mean and when it is worth changing the model.
In short
- If you do not know what to pick, leave “Automatic” or whatever is marked Recommended. For most jobs that is enough.
- Change the model when you have a reason. The task is too hard: take a stronger one. The task is simple and you want it quicker and cheaper: take a lighter one.
- The strongest is not always the best. It is usually slower and uses up your subscription limit faster.
What you are choosing between
| What | What it means |
|---|---|
| Quality | A stronger model keeps a big project in mind better and slips less on tricky logic. |
| Speed | A lighter model answers faster. For small edits the difference is noticeable. |
| Limit | A stronger model usually uses your subscription limit faster. If your limit runs out early, check whether the strongest model is sitting on easy work. |
How to read the names
A name is a family plus a version. In “Opus 5.5”, Opus is the family (the power level) and 5.5 is the version. A newer version of the same family is usually better than an older one. Old versions stay so that work can be reproduced, and SHIPIT tucks them under More models.
The pane settings window has two more rows that are easy to miss:
- Automatic: the tool picks the model itself.
- Or type a model name: a field for a name that is not in the list. Type it exactly: with a typo the agent will not start or will answer with an error. If you can choose from the list, choose from the list.
What each tool offers
Claude Code
| Family | What it is for |
|---|---|
| Fable, Opus | The strongest level: big and tricky tasks, a hard bug, architecture, a large rework. Slower, and they use the limit faster. |
| Sonnet | Everyday work: a new feature, a fix, tests. A good balance of power and speed. |
| Haiku | The lightest and fastest: explain code, a small edit, a quick question. |
Codex
The descriptions below come from Codex’s own catalog. Version names change, so only the families are listed here.
| Family | What it is for |
|---|---|
| Astra | The strongest: the most demanding work. |
| Sol | The workhorse: coding and everyday work. |
| Terra | A balanced model for more straightforward work. |
| Luna | Fast and affordable, for easier tasks. |
Other tools
- Gemini: the auto row leaves the choice to the tool, Flash is faster and lighter, Pro is stronger.
- Grok: one main model row; you set the depth of thinking with effort (below).
- OpenCode: can use models from many providers, so the list depends on what you connected.
- Aider, terminal, browser, Markdown: there is no model to choose here, or the choice does not apply to the pane.
Effort: how much the model thinks
Many agents have an effort setting: Low, Medium, High, Extra high, Max. It tells the model how much to reason before it answers. More effort gives a more accurate answer on hard problems, but it is slower and costs more of your limit. For simple work, Low or Medium is enough.
The Grok CLI describes its levels like this: Low is the fastest responses, best for simple tasks; High is thorough reasoning and recommended; Extra high is maximum reasoning for the hardest tasks.
What to pick for what
| What you are doing | Which level | Examples |
|---|---|---|
| Explain code, a small edit, a rename, a question | Light | Haiku, Luna, Gemini Flash |
| Regular development: a feature, a fix, tests | Middle | Sonnet, Sol, Terra |
| A hard bug, architecture, a large rework, security | Strongest, higher effort | Fable, Opus, Astra |
| Not sure | Whatever the tool picks | Automatic or Recommended |
How to change the model in SHIPIT
- Open the settings of the pane whose model you are changing.
- In the Model list choose the one you want. If it is not there, open More models or type the name in the field below.
- Press Save & restart. Nothing changes until you save and restart the pane.
Common mistakes
- Always the strongest. The limit runs out fast, and for simple work there is almost no difference. Keep the strong model for hard work.
- A model name with a typo. The agent will not start. Choose from the list.
- Effort at maximum for everything. Answers get slower with no gain on easy tasks.
Models and names change quickly. If SHIPIT shows a model that is not listed here, look at its level: stronger, balanced or light. As of October 2026.