Quick Answer
GPT-6 Sol and GPT-6 Luna are OpenAI’s new lower-cost models for coding, agentic work, and high-volume tasks. OpenAI launched both models on September 22, 2026, with API prices it says are 50% below corresponding GPT-5.6 promotional rates. Developers can use the APIs now, while ChatGPT access is gradual and depends on the plan and product.
Key Takeaways
- GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens.
- GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens.
- Both GPT-6 models support text, image input, multilingual use, vision, and a 1.05 million-token context window.
- GPT-6 Sol is intended for complex coding and agentic workflows, while GPT-6 Luna targets focused high-volume work.
- OpenAI classifies both models as High capability, below Critical, in cybersecurity and biological-and-chemical domains.
What did OpenAI launch with GPT-6 Sol and Luna?
GPT-6 Sol and GPT-6 Luna are two new OpenAI models designed to offer faster and lower-cost access than their GPT-5.6 counterparts. OpenAI launched the models on September 22, 2026, and said both were built using methods similar to those behind GPT-6 Astra. OpenAI’s launch announcement identifies Sol as the model for more complex coding and agentic workflows, while Luna is positioned for narrower, high-volume tasks.
The distinction matters because model selection is increasingly a cost and workflow decision rather than a simple choice between a capable model and a lightweight one. GPT-6 Sol is intended for tasks that require more involved reasoning across a coding or agent workflow. GPT-6 Luna is intended for repeatable work where lower cost and throughput matter more than using the highest-capability option for every request.
OpenAI’s release is separate from recent competing AI model launches, including Claude’s lower-cost API update and other models that emphasize larger context windows or reasoning controls. The practical decision for developers is to identify which requests actually need complex orchestration before assigning them to Sol rather than Luna.
How much do GPT-6 Sol and Luna cost through the API?
GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, while GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens. OpenAI said the new rates are 50% lower than the corresponding GPT-5.6 promotional prices. That reduction changes the economics for applications that process large prompt volumes, generate lengthy responses, or run repeated automated tasks.
| Model | Input price per 1 million tokens | Output price per 1 million tokens | OpenAI’s stated use case |
|---|---|---|---|
| GPT-6 Sol | $2 | $10 | Complex coding and agentic workflows |
| GPT-6 Luna | $0.10 | $0.50 | Focused, high-volume tasks |
| GPT-5.6 Sol promotional pricing | $4 | $20 | Previous comparison point |
| GPT-5.6 Luna promotional pricing | $0.20 | $1.20 | Previous comparison point |
GPT-6 Sol’s pricing is most relevant when an application needs its more advanced workflow focus, because output tokens cost substantially more than input tokens. GPT-6 Luna’s pricing makes more sense for large volumes of focused requests, especially where an application can keep prompts and responses concise. Developers should measure both input and output usage before estimating savings, because the bill depends on the amount of each token type a workload produces.
OpenAI lists the available models and their technical details in its model documentation. The sensible approach is to test representative production prompts with both models before changing a default model across an application.
Where are GPT-6 Sol and Luna available?
GPT-6 Sol and GPT-6 Luna are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, although OpenAI said the rollout would happen gradually. GPT-6 Luna is also available to Free and Go users in the desktop app. OpenAI said the models are not yet available in the regular ChatGPT chat experience, which means access can differ even among users who have a paid ChatGPT plan.
ChatGPT availability is therefore not the same as general API availability. A developer can use a model through an API integration while a regular ChatGPT user may not see that model in the standard chat interface. OpenAI’s ChatGPT release notes provide the company’s ongoing product availability information, but users should expect feature access to vary during a gradual rollout.
The limitation is important for teams that want to standardize an internal workflow immediately. A team should confirm whether its members use ChatGPT Work, Codex, the desktop app, or the regular chat experience before writing documentation around a model option that may not yet appear for every user. Organizations that already use AI in shared work tools may also want to consider how AI features in Microsoft Word fit alongside a separate coding or API workflow.
What capabilities do GPT-6 Sol and Luna share?
GPT-6 Sol and GPT-6 Luna both accept text and image input, generate text, support multilingual use and vision, and provide a 1.05 million-token context window. A context window is the amount of information a model can consider within a request and its conversation history. The 1.05 million-token limit allows developers to provide substantially more reference material in a single workflow than smaller-context models allow.
The shared capability set means a developer does not need to move to Sol solely to obtain image input, text output, multilingual support, or a large context window. The main choice remains the complexity and volume of the task. A focused extraction, classification, or repetitive processing workflow may fit Luna, while a more involved coding system may justify Sol.
A large context window does not guarantee that every long prompt produces the best answer. Large inputs still require careful document selection, clear instructions, and checks for incorrect or incomplete output. The practical action is to treat the context window as capacity, not as a substitute for application design, evaluation, or human review in high-impact work.
How do GPT-6 Sol and Luna differ in practical use?
GPT-6 Sol is OpenAI’s option for complex coding and agentic workflows, while GPT-6 Luna is OpenAI’s option for focused, high-volume tasks. An agentic workflow is a system in which a model carries out multiple steps toward a goal, such as reviewing information, writing code, and passing results to another tool. The distinction gives developers a way to align a model’s cost with the difficulty of the work.
GPT-6 Sol is the stronger fit when an application needs a model to manage a more complex sequence of coding or tool-based tasks. GPT-6 Luna is the more economical fit when a system handles many well-defined requests that do not need that level of workflow complexity. OpenAI’s positioning does not mean Luna is unsuitable for all coding or Sol is necessary for every agent, so teams should validate their own use cases.
The practical response is to separate requests into at least 2 groups: complex tasks that need advanced orchestration and routine tasks that need reliable throughput. Companies building systems with multiple specialized AI workers can apply the same routing principle used in multi-agent support workflows, assigning each request to the least expensive model that meets the quality requirement.
What do OpenAI’s safety classifications mean for GPT-6 Sol and Luna?
GPT-6 Sol and GPT-6 Luna are classified by OpenAI as High capability, but below Critical, in both cybersecurity and biological-and-chemical domains. OpenAI also reported that, in an internal simulated Codex deployment across 50,319 tasks, GPT-6 Sol generated 42 severity-3-or-higher misalignment flags, compared with 66 for GPT-5.6 Sol. OpenAI’s safety documentation provides the company’s published assessment material.
OpenAI’s classification is a company assessment, not an independent guarantee that a model cannot be misused or make harmful errors. The reported decline in internal flags is useful context, but the result comes from OpenAI’s own simulated deployment and should be treated as self-reported. Security teams should continue to apply access controls, logging, review processes, and limits on tools that can take consequential actions.
The most important safety decision is to avoid giving any AI system broad permissions without oversight. Developers should test what a model can access, what it can send, and whether a human approval step is required before an external action occurs. Teams handling credentials, customer records, financial data, or production infrastructure should stop and involve their security or compliance staff before deploying an autonomous workflow.
What should developers do before moving workloads to GPT-6 Sol or Luna?
Developers should compare GPT-6 Sol and GPT-6 Luna against representative tasks before moving a production workload. Start with a small evaluation set containing the prompts, files, images, expected outputs, and edge cases that matter to the application. This approach reveals whether Sol’s complex-workflow focus produces enough additional value to justify its higher rate.
- Identify the 2 main workload types, such as complex coding tasks and high-volume focused requests.
- Measure input and output token use for each workload before estimating API costs.
- Test text, image, multilingual, and long-context requests separately if the application uses those capabilities.
- Review incorrect outputs, incomplete responses, and unsafe tool actions before changing the production default.
- Limit model permissions and require human approval for consequential external actions.
GPT-6 Luna is a reasonable starting point for a focused workload where cost and scale are the primary requirements. GPT-6 Sol is a reasonable candidate when a workflow genuinely depends on complex coding or agentic behavior. Developers should not treat either model as a replacement for testing, because a model’s stated use case does not establish performance for a particular company’s data or process.
The stop line is clear for regulated, security-sensitive, or safety-critical systems. Do not deploy GPT-6 Sol or GPT-6 Luna with unrestricted access to customer data, production systems, money movement, or security tools until the organization’s security, privacy, and legal reviewers approve the design.
FAQ
What are GPT-6 Sol and GPT-6 Luna?
GPT-6 Sol and GPT-6 Luna are OpenAI models launched on September 22, 2026, for complex coding and agentic workflows or focused high-volume tasks. Both models accept text and image input, generate text, support multilingual use and vision, and have 1.05 million-token context windows.
How much does GPT-6 Sol cost?
GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens. OpenAI said those rates are 50% lower than the corresponding GPT-5.6 Sol promotional prices of $4 and $20.
How much does GPT-6 Luna cost?
GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens. OpenAI previously listed GPT-5.6 Luna promotional pricing at $0.20 for input and $1.20 for output per million tokens.
Can free ChatGPT users access GPT-6 Luna?
GPT-6 Luna is available to Free and Go users in the ChatGPT desktop app. OpenAI said the broader rollout is gradual, and GPT-6 Sol and Luna are not yet available in the regular ChatGPT chat experience.
Are GPT-6 Sol and Luna safe for autonomous work?
GPT-6 Sol and GPT-6 Luna have High capability classifications in cybersecurity and biological-and-chemical domains, according to OpenAI, so autonomous use needs controls. Organizations should restrict permissions and require human approval before a model takes consequential external actions.