GPT-6 is OpenAI's family of AI models that includes Astra, Sol, and Luna. They can reason through problems, generate text and code, interpret images, and work with connected tools. Astra targets the most demanding work, Sol balances capability and cost, and Luna targets focused tasks at high volume.
Copy link to headingWhich models are in the GPT-6 family?
The model names identify different choices within the same generation. OpenAI's model overview describes their intended roles:
These roles describe where to start. Sol and Luna have different prices but share support for image input and function calling, along with the same context-window size. Both can analyze images and request actions through connected tools.
Those shared capabilities do not guarantee equal results on a task, especially when completing it requires interpreting conflicting information or making several dependent decisions.
Copy link to headingWhat changed with GPT-6 Sol and Luna?
OpenAI's Sol and Luna announcement extends the GPT-6 family with lower-cost alternatives to Astra. OpenAI reports improvements over the models' GPT-5.6 counterparts in areas including coding and factuality. It also describes changes to communication, with shorter answers and less jargon.
For developers, the release also changes the economics of repeated context. OpenAI describes improved default cache hit rates and more control over which opening portions of a prompt get cached. Reusing a long instruction set or document collection can reduce the amount of input that needs fresh processing.
Copy link to headingWhat can GPT-6 do in an application?
You can ask the model for different outputs depending on the input:
For a report, a summary with supporting passages.
For a screenshot, an explanation of a visible error.
For code, a proposed change with a description of the affected behavior.
The input determines what evidence the model has. Pasting a support ticket supplies the customer's account of a problem. Connecting an order-lookup tool lets the assistant consult a record. Those two sources can disagree, so a useful answer should identify which source supports each statement.
GPT-6 Astra supports work including research and computer use. Sol and Luna also accept text and images and generate text responses. An image can supply evidence about a screen, while an execution tool gives the application a way to act on what the model identifies.
Copy link to headingWhat does reasoning mean in GPT-6?
GPT-6 can use reasoning tokens to work through a problem before answering. The reasoning effort setting lets you adjust how much reasoning the model uses. Lower effort can reduce response time and token costs. Higher effort allows more reasoning for difficult tasks, but can take longer and cost more.
That setting is separate from the answer length you want. Producing a short recommendation can require substantial analysis, whereas a long response might largely repeat information already provided. You can ask for a concise answer while allowing the model more reasoning effort.
Extracting a date from a document may require less reasoning than resolving conflicting policies. The model needs both policies and a basis for deciding which takes priority, such as their effective dates or an explicit precedence rule. Higher reasoning effort can help it compare the available evidence but cannot compensate for a missing policy.
Copy link to headingWhat does a large context window let you do?
The context window limits how much material a model can work with in a request. OpenAI's key concepts guide explains tokens as the units used to represent model input and output. Token counts differ from word counts, so a context limit should not be read as a fixed number of document pages.
The GPT-6 model comparison lists a 1,050,000-token context window and a 128,000-token maximum output for Astra, Sol, and Luna. Input, reasoning, and generated output must fit within the context window. The output allowance includes reasoning tokens as well as the visible answer, so 128,000 tokens is not a guarantee of that much visible text.
Larger context windows let you supply a project's requirements alongside its implementation notes so the model can compare the plan with the implementation. Preserve document titles and dates to distinguish current decisions from earlier versions.
Select documents for the question you're asking, even when more material would fit. Notes from unrelated projects can introduce conflicting information and make relevant evidence harder to identify. Leave enough of the context budget for reasoning and the response.
Providing up-to-date documents allows the model to use information that may not have been available during training. Providing that context does not, by itself, update the model's trained knowledge.
Copy link to headingHow does GPT-6 support work across multiple steps?
Agents let a model use tool results to decide what to do next. During a coding task, the agent might inspect a file, propose a change, and run a check before responding. The integration connects the model to an execution environment and supplies the operation results.
Asynchronous tool calling lets GPT-6 continue independent work while an application runs a tool. An assistant could draft a report's outline during a data lookup, then use the results to write the sections that depend on them.
Mid-turn steering lets users add or change requirements before a response finishes. Your application sends the update through a WebSocket connection to the Responses API and handles any pending tool results. Steering does not undo earlier actions or cancel tools already in progress.
Copy link to headingHow do you access GPT-6?
You can use GPT-6 through an existing app or integrate it into your own software:
ChatGPT Work and Codex let you use the models without building an integration. Sol and Luna are available to Plus, Pro, Business, Enterprise, and Edu users. Free and Go users can access Luna in the desktop app. Sol and Luna are not available in Chat.
The OpenAI API gives your application direct access to the models. Use the Responses API for model requests and tool workflows, with
gpt-6-solorgpt-6-lunaas the model identifier.AI Gateway provides one endpoint for calling models from multiple providers, with request logs and spending information. Use
openai/gpt-6-soloropenai/gpt-6-lunafor these models. Your application can use Gateway even when hosted outside Vercel.
Follow the setup guide to install the AI SDK and send your first request. Vercel deployments can use OIDC authentication without managing an API key. You can also authenticate by adding the AI_GATEWAY_API_KEY environment variable.
Copy link to headingWhat are GPT-6's limits?
Generated answers can omit evidence or draw conclusions that the source material does not support. Citations let readers inspect the original passage, but that passage still needs to substantiate the claim. Questions the available material cannot answer need a different response: the application should request the missing information or explain what remains unknown.
The tools you connect determine what information the model can retrieve and which actions it can request. An order-status assistant needs a lookup function with access to the relevant records. Your application executes the lookup requested through function calling and returns the result for the model to use in its answer.
Sol and Luna do not natively accept or generate audio and video. Applications that work with those formats need an additional model or processing step to handle them.
Copy link to headingFrequently asked questions
Copy link to headingDoes GPT-6 include Astra, Sol, and Luna?
Yes. The GPT-6 family includes Astra, Sol, and Luna. OpenAI positions Astra for its most demanding work, Sol for complex coding and agent workflows, and Luna for focused tasks at high volume.
Copy link to headingIs GPT-6 the same thing as ChatGPT?
No. GPT-6 names a model family, while ChatGPT is an application through which people can use OpenAI models. Developers can also integrate GPT-6 into their own products through an API.
Copy link to headingIs GPT-6 only for coding?
No. OpenAI also describes GPT-6 uses in professional work and research. Coding is one application of the family, and the right model depends on the task.
Copy link to headingCan I use GPT-6 Sol and Luna through Vercel?
Yes. You can access both models through AI Gateway. Use the Gateway model identifiers openai/gpt-6-sol and openai/gpt-6-luna when configuring requests.