Documentation
Which ChatGPT model to choose?
Last updated:
OpenAI models can differ in price by several times, yet on simple tasks a cheap model answers about as well as the flagship. So there's no point paying for the most powerful model everywhere. Below we go through which model to use for which task.
Pick a model by task
Start with the recommended model and step up only if quality falls short. That way you pay for power only where it's genuinely needed.
| Task | Model | Why |
|---|---|---|
| Autocomplete, classification, field extraction | gpt-5.4-mini | Cheap and fast, quality is enough |
| High-volume simple requests, drafts | gpt-5.6-luna | Cheapest in the new lineup |
| Everyday code, writing, analysis | gpt-5.6-terra | The default, balanced price and quality |
| Hard reasoning, large refactors | gpt-5.6-sol | Flagship, top quality |
| Agent loops and planning | gpt-5.6-sol | Holds long chains of steps better |
| Previous generation, stability | gpt-5.5 / gpt-5.4 | Proven models at the same discount |
Models and pricing
Every rate is 50% below official OpenAI. Prices are per 1M tokens via ChatGPT API.
When to use
Hard reasoning, large code, agents
When to use
Everyday code, writing, analysis
When to use
High-volume simple requests, drafts
When to use
Solid all-rounder
When to use
Good price-to-quality
When to use
Autocomplete, classification, extraction
How to switch models
The model goes in the standard `model` field of the request, exactly as with the official OpenAI API. The key and base URL don't change, so switching is a one-liner.
// Same key, same base URL — only the model changes.
model: "gpt-5.4-mini" // cheapest, high volume
model: "gpt-5.6-luna" // fast and cheap
model: "gpt-5.6-terra" // balanced default
model: "gpt-5.6-sol" // flagship, hard tasksFrequently asked questions
Which ChatGPT model is the cheapest?+
In the new lineup it's gpt-5.6-luna; in the previous generation, gpt-5.4-mini. Both suit high-volume simple tasks. All prices are 50% below official.
Which model is best for coding?+
For the hardest code, refactoring and agents, gpt-5.6-sol. Everyday development is fine on gpt-5.6-terra, and bulk completions are cheaper on gpt-5.6-luna.
How does gpt-5.6-sol differ from terra?+
sol is the flagship: stronger on hard reasoning and long agent chains, but more expensive. terra is cheaper and handles most everyday tasks with no noticeable quality drop.
Can I use different models in one project?+
Yes. One key and base URL cover all models, so you can route simple requests to a cheap model and hard ones to the flagship by changing only the model field.