GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna: what the names mean
Astra is OpenAI’s option for its most demanding reasoning tasks; Sol targets a balance of capability and cost; Luna is intended for focused work at high volume. A higher-looking version number does not turn these into a simple best-to-worst ranking. The name also does not tell you whether your ChatGPT plan, workspace or API project can use the model.
Source notes
This is an AI-assisted Ithy explanation, based on OpenAI documentation checked on October 25, 2026 in New York. It is not an independent performance benchmark or an OpenAI announcement. Product access and documentation can change. Follow the official links below for your actual app and account.
Start with the role, not the number
OpenAI’s model catalogue presents three different starting points. Its descriptions are the vendor’s positioning, rather than a guarantee about every task.
| Name | API identifier | Role described by OpenAI |
|---|---|---|
| GPT-6 Astra | gpt-6-astra | Difficult reasoning, coding and other demanding work. |
| GPT-6.1 Sol | gpt-6.1-sol | A balance between complex-work capability and cost. |
| GPT-6 Luna | gpt-6-luna | Efficient handling of focused, repeated tasks. |
For example, reviewing a long document for conflicting requirements and extracting a short label from many similar messages are different jobs. Those examples illustrate why separate model tiers can make sense; they are not test results showing that one model wins.
What is GPT-6 Astra?
The Astra model page describes uses including reasoning, software work, research and document creation. Its documented API inputs include text and images, with text output. The page also lists tools a compatible application can provide. A model’s tool support does not establish that a particular app has enabled that tool or given it permission to act.
If you want a current, sourced answer, check whether the app actually used search and opened relevant material. The model name alone is not evidence that it looked something up.
Why can Sol be called 6.1 while Astra is called 6?
The names identify different offerings. OpenAI’s Sol page positions it near Astra for complex tasks at a lower cost, and recommends comparing the two on the work you need to do. That is a vendor description, not proof that the models are interchangeable. Our practical suggestion is to compare a few representative tasks before assuming that the version number settles the choice.
What is Luna for?
The Luna page emphasizes efficiency for focused, high-volume work. A sensible evaluation might include short summaries, extracting specified fields or applying a clear classification rule. Check errors as well as speed: a low-cost answer that needs extensive correction may not be the cheaper workflow.
ChatGPT, Codex and API access are separate checks
OpenAI’s workspace availability guidance says access depends on the product surface and sign-in boundary. Workspace settings do not automatically grant the same access in every Codex client or API project. In eligible Enterprise and Edu workspaces, Astra starts disabled until an owner enables it; a local model setting does not override that control.
For an individual ChatGPT plan, check the models actually offered in your current app and plan. An API catalogue entry does not by itself establish personal-plan eligibility, an allowance or a subscription price. Likewise, being able to select a name is separate from permission to read files, use tools or access another service.
A large context window does not mean unlimited uploads or live knowledge
The three model pages list a 1,050,000-token context window, with separate input and output limits. An app can impose a smaller upload, prompt or output allowance. A model limit therefore cannot be used as a promise about a particular product.
The documented knowledge cutoffs are April 30, 2026 for Astra and Sol, and May 18, 2026 for Luna. A later date on a conversation does not make the model’s built-in knowledge current. When newer information matters, inspect the retrieval tools and sources the application actually used.
A note for people building an app
Check the endpoint as well as the model. Sol’s page directs tool-calling applications to Responses; its Chat Completions support excludes tool calls. Luna’s page allows Chat Completions function calling only with reasoning effort set to none. Copying one model’s settings into another request can therefore change what works.
API charges also depend on the current pricing schedule and request configuration. Read the actual model page and pricing terms rather than treating a model name, context limit or ChatGPT subscription as a dollar quote.
A quick way to choose without guessing
- Identify the app and account you are using.
- Check which model is actually available there.
- Try the same small set of representative tasks, including cases where an error would matter.
- Inspect sources, unsupported claims and needed corrections.
- Compare total time and applicable cost, including any tools and review work.
This is a decision method, not a claim that Ithy ran a new three-model benchmark. For Ithy’s own current research workflow, see how Ithy works; this explanation does not imply that Ithy runs Astra.