GPT-6 Astra, Sol and Luna: which to choose?
Compare GPT-6 Astra, Sol and Luna by use case, API price and availability in ChatGPT, Work and Codex. Plus, why Terra belongs to GPT-5.6.
GPT-6 at a glance: three models, not four
OpenAI's GPT-6 lineup consists of Astra, Sol and Luna. Astra targets the hardest work, Sol balances capability and cost, and Luna handles focused tasks at scale. There is no GPT-6 Terra in the official catalog checked on September 26, 2026: Terra belongs to the previous GPT-5.6 generation.
Choosing a model is not just about a version number. The useful question is how difficult the task is, how much quality matters, and what time and cost it warrants. The largest model is not always the best starting point.
| Model | Best role | Concrete example |
|---|---|---|
| GPT-6 Astra | The hardest reasoning and development work | Audit an entire project, connect sources and verify fixes |
| GPT-6 Sol | Balanced coding and multi-step workflows | Build a tool, analyze data and produce a usable report |
| GPT-6 Luna | Efficient, well-scoped tasks at scale | Classify queries or extract structured fields |
Astra: for difficult end-to-end work
OpenAI describes GPT-6 Astra as its flagship model for complex reasoning and coding. It is most useful when a task involves linked decisions: understanding constraints, inspecting files or sources, using tools, producing a result and checking it afterward.
For SEO, Astra can connect a technical audit with indexed pages, existing content and site structure before proposing changes. For software, it can examine a codebase, fix several modules and check side effects. That does not make its answers infallible: recent facts and published changes still need verification.
Sol: the starting point for many substantial tasks
GPT-6 Sol sits in the middle of the lineup. OpenAI positions it for coding and agentic workflows, at a lower API price than Astra. A practical approach is to start with Sol for serious everyday work and move to Astra when complexity or the cost of a mistake justifies it.
Examples include building and testing a focused feature, analyzing exports, preparing a marketing report from several sources or revising a substantial document. For a simple, repetitive job, even Sol may be more than you need.
Luna: handle volume without paying for the flagship
GPT-6 Luna has the lowest API price of the three. It fits stable instructions with a clearly defined output: keyword categorization, information extraction, title normalization or a first pass over a large batch of documents.
One useful pattern is to let Luna handle straightforward cases and send ambiguous ones to Sol or Astra. Check the error rate on real examples before scaling up; a cheap model is not a saving if every result needs to be redone.
Where does GPT-5.6 Terra fit?
Terra is not a fourth GPT-6 model. GPT-5.6 Terra was the balanced model in the previous family, alongside GPT-5.6 Sol and Luna. In GPT-6, Sol occupies the middle position and Astra is the flagship.
Model names should not be compared on their own across generations. GPT-5.6 Sol and GPT-6 Sol have different positions in their respective lineups. See our GPT-5.6 Sol, Terra and Luna guide for the earlier generation.
Where are these models available?
Availability depends on the product, plan and sometimes workspace settings. The OpenAI API, a standard ChatGPT conversation, ChatGPT Work and Codex do not necessarily show the same model picker.
- OpenAI API: the official model IDs are
gpt-6-astra,gpt-6-solandgpt-6-luna. The per-token prices below refer to this product. - ChatGPT Work and Codex: Astra, Sol and Luna may be offered depending on your plan, workspace settings and rollout. Sol and Luna are not selectable in regular ChatGPT conversations.
- Regular ChatGPT: GPT-6 Pro, powered by Astra, is available on the eligible plans listed by OpenAI. Astra access in Work or Codex on another plan does not imply the same option is present in Chat.
For the tools discussed on this site, explore our ChatGPT Pro page and Rankerfox AI tools overview. To see which model you can select right now, check the picker in your own session.
API prices: a reference point, not a subscription price
As of September 26, 2026, OpenAI's model catalog lists these standard per-million-token input and output API prices. They are not ChatGPT subscription prices or the price of access through Rankerfox.
| Model | API input | API output |
|---|---|---|
| Astra | $10 | $50 |
| Sol | $2 | $10 |
| Luna | $0.10 | $0.50 |
Actual billing can also depend on tools, caching, context size and processing options. For a precise budget, consult current API pricing. The difference between API pricing and subscription pricing is also illustrated by ToolsBoxSEO's French-language analysis of a possible Pro Max plan: its reported price alone cannot establish Codex quotas or model access.
Which model should you choose?
- Luna when the instructions are clear, examples are plentiful and results are easy to check.
- Sol when the work involves analysis, code, tools or connected steps.
- Astra when the task is difficult, ambiguous, lengthy or expensive to get wrong.
The strongest test is your own set of examples: run the same cases through two models, measure corrections and compare total time. A cheaper model per token can cost more in human rework, and the reverse can also be true.
Frequently asked questions
Is there a GPT-6 Terra?
Not in the current official lineup. Terra belongs to GPT-5.6. OpenAI highlights Astra, Sol and Luna for GPT-6.
Is GPT-6 Sol the same as GPT-5.6 Sol?
No. GPT-5.6 Sol was its family's flagship; GPT-6 Sol is the balanced model in the new lineup.
Can I select Sol or Luna in a regular ChatGPT conversation?
OpenAI says no: those models are for eligible Work and Codex access, as well as the API. Availability may change; the product's model picker is the easiest practical check.
Official sources
Information checked against these OpenAI pages on September 26, 2026: