Previous generation
GPT-5.6
GPT-5.6 is the generation that came before GPT-6. It is still widely used, mostly in apps and workflows that were built and tested on it, and it is sold in three variants.
At a glance
Specifications
- Family
- GPT-5.6
- Previous generation, three variants
- Released
- July 9, 2026
- Context window
- 1.05M
- Maximum tokens in one request
- Input price
- From $0.20
- Per 1 million tokens
- Output price
- From $1.20
- Per 1 million tokens
- A heavy month
- $1
- 2M input and 500K output tokens at API prices
Not the model in the free chat
What it is
GPT-5.6 came out about two months before the GPT-6 family. Like GPT-6, it accepts about 1.05 million tokens in one request.
It comes in three variants with different prices: Sol, Terra and Luna. The standard GPT-5.6 listing has the same price as the Luna variant, the lightest of the three.
For new projects, GPT-6 is usually the better starting point. GPT-6 Luna costs less per token than GPT-5.6, and GPT-6 Sol costs the same as GPT-5.6 Sol, so moving to the newer generation rarely raises the bill. GPT-5.6 remains useful when you need answers to behave exactly as they did before.
What it does well
Known behaviour
Many teams have already tested their prompts on GPT-5.6. Staying on it avoids surprises while you evaluate GPT-6.
Large context
Reads about 1.05 million tokens per request, the same as the GPT-6 models.
Choice of variants
Three price points let you match the variant to the task.
Still capable
Handles everyday writing, coding and questions well. GPT-6 is a step forward, not a different kind of tool.
Limits and when to pick another model
- It is the older generation. For new work, the GPT-6 models are the natural first choice.
- Even its lightest variant costs more per token than GPT-6 Luna.
- GPT-5.6 Sol and GPT-6 Sol have the same list price, so the older model saves nothing at that level.
- Providers retire older models over time. If you depend on GPT-5.6, plan a move to GPT-6.
- It is not the model in this site's free chat, which runs GPT-6 Luna.
Pick another model when
For a new assistant or app, start with GPT-6 Luna for speed and cost, or GPT-6 Sol for depth. Keep GPT-5.6 where your prompts are already tested and the output must stay consistent until you have compared it with GPT-6.
Best tasks, with example prompts
These are the jobs where this model gives the best return for its speed and price.
- Running an existing app while you test GPT-6 side by side
- Keeping outputs consistent in a workflow that was tuned on GPT-5.6
- Comparing the old and new generation on your own prompts
- Everyday writing and questions in tools that only offer GPT-5.6
Example prompts. You can copy them or open them in the free chat, which answers with GPT-6 Luna.
I run a customer support assistant on GPT-5.6. Write a simple test plan to compare it with GPT-6 Luna before switching: what to measure, how many sample questions to use and how to decide.
Suggest 20 test questions for a support assistant in an online shop, covering easy, tricky and off-topic cases. For each one, say what a good answer must include.
I will paste two answers to the same question, one from an older model and one from a newer one. Compare them on accuracy, clarity and tone, and say which one you would send to a customer and why. [paste both answers]
Variants and prices
OpenRouter lists three GPT-5.6 variants. Sol and Terra cost roughly ten times as much as Luna, the lightest variant. Terra has the same input price as Sol and a higher output price.
| Variant | Input per 1M | Output per 1M | A heavy month |
|---|---|---|---|
| GPT-5.6 Sol | $2 | $10 | $9 |
| GPT-5.6 Terra | $2 | $12 | $10 |
| GPT-5.6 Luna | $0.20 | $1.20 | $1 |
If you are choosing a variant for a new project, first compare it with the GPT-6 model at a similar price.
How it compares with other GPT models
In its standard form, GPT-5.6 is priced between GPT-6 Luna and GPT-6 Sol. At the Sol level the price is the same across both generations, and at the light end GPT-6 Luna is cheaper. In most cases the newer family gives you more for the same money.
| Model | Released | Context window | Input price | Output price | Speed | Depth |
|---|---|---|---|---|---|---|
| GPT-6 Astra | September 4, 2026 | 1.05M | $10 | $50 | ||
| GPT-6 Sol | September 22, 2026 | 1.05M | $2 | $10 | ||
| GPT-6 Luna | September 22, 2026 | 1.05M | $0.10 | $0.50 | ||
| GPT-5.6This page | July 9, 2026 | 1.05M | $0.20 | $1.20 |
All four current GPT models side by side. Speed and depth are relative ratings within this lineup, not benchmark scores.
Other models
- GPT-6 AstraThe top GPT-6 model, for hard multi-step problems where accuracy matters more than speed or cost.$10 / $50
- GPT-6 SolThe sensible default for serious work: clearly deeper than Luna, much faster and cheaper than Astra.$2 / $10
- GPT-6 LunaQuick, light and inexpensive. Good for everyday questions, summaries and bulk work, and the model in this site's free chat.$0.10 / $0.50
- ChatGPT alternativesHow the GPT-6 models compare with Claude, Gemini, Grok and DeepSeek
Questions
Is GPT-5.6 still available?
Yes. As of 25 September 2026 it is listed on OpenRouter with three variants. Older models are usually retired at some point, so check with your provider before you rely on it for the long term.
Should I switch from GPT-5.6 to GPT-6?
For most uses, yes, after a test. GPT-6 Luna is cheaper than the standard GPT-5.6 listing, and GPT-6 Sol costs the same as GPT-5.6 Sol. Run your own prompts through both before you switch, so there are no surprises.
What do Sol, Terra and Luna mean for GPT-5.6?
They are variants at different price points. Luna is the lightest and cheapest, while Sol and Terra cost roughly ten times as much. The table on this page lists their prices.
How much does GPT-5.6 cost?
GPT-5.6 is listed on OpenRouter at $0.20 per million input tokens and $1.20 per million output tokens. A heavy month of 2 million input and 500,000 output tokens would cost about $1. Using the free chat on this site costs you nothing.
Test your prompt on GPT-6 Luna
Paste a prompt you use with GPT-5.6 into the free chat and compare the answer.