As of November 9, 2026, OpenAI’s API catalog lists the GPT-6 family—not a newly announced successor such as GPT-7. It identifies GPT-6 Astra (gpt-6-astra) as the flagship for complex reasoning and coding. OpenAI positions GPT-6.1 Sol as a capability-and-cost balance, GPT-6.1 Terra for visual reasoning and computer-use tasks, and GPT-6 Luna for focused, high-volume work. These are catalog descriptions, not a guarantee that a particular model will perform best for your application. (OpenAI API model catalog)
If by “next GPT model” you mean what comes after GPT-6, the official materials cited here do not announce a later major GPT generation or a release date. Astra is the family’s flagship; Sol, Terra, and Luna are other options in that family. (OpenAI API model catalog)
GPT-6 API models at a glance
The catalog lists these standard input and output prices per 1 million tokens (MTok). It also lists a 1.05 million-token context window and a 128,000-token maximum output for each of the four models. (OpenAI API model catalog)
| Model | API model ID | Catalog positioning | Input / output price per 1M tokens |
|---|---|---|---|
| GPT-6 Astra | gpt-6-astra |
Most demanding work; complex reasoning and coding | $10 / $50 |
| GPT-6.1 Sol | gpt-6.1-sol |
Complex work with a balance of capability and cost | $2 / $10 |
| GPT-6.1 Terra | gpt-6.1-terra |
Visual reasoning, precise computer use, and writing | $1 / $5 |
| GPT-6 Luna | gpt-6-luna |
Focused, cost-sensitive, high-volume tasks | $0.10 / $0.50 |
These qualitative descriptions are OpenAI’s stated positioning, not independent comparisons. For example, OpenAI describes Sol as near-Astra performance for complex work at lower cost; the catalog capture does not provide workload-specific benchmarks to establish how the models compare on your task. (OpenAI API model catalog)
A simple baseline token-cost example
For an illustrative request with 100,000 input tokens and 20,000 output tokens, applying the catalog’s listed standard rates gives these baseline calculations:
- Astra: $1 input + $1 output = $2
- Sol: $0.20 input + $0.20 output = $0.40
- Terra: $0.10 input + $0.10 output = $0.20
- Luna: $0.01 input + $0.01 output = $0.02
These are baseline arithmetic examples, not estimates of a particular bill. They assume the listed standard token rates and exclude cached tokens, tool use, and any other applicable pricing adjustments. Check the Astra model page and the relevant model’s current pricing details before budgeting; some tool calls may carry separate fees.
Which model should you try first?
- Try Astra when the task is demanding—for example, complex coding, research, or document work. Its higher listed price may be worthwhile if your own evaluation shows a meaningful improvement.
- Try Sol when you need capability for complex work but want a lower listed token price. Test it against Astra on your own tasks rather than assuming “near-Astra” positioning guarantees equal results. (OpenAI API model catalog)
- Try Terra when the workflow centers on visual reasoning, computer use, or clear writing. OpenAI describes it as a fast, efficient model for those tasks; test whether that positioning fits your inputs and required actions. (OpenAI API model catalog)
- Try Luna for repeated, focused tasks at high volume where low token cost matters. Check that it meets your quality and formatting requirements before routing production traffic to it. (OpenAI API model catalog)
A practical evaluation is to run each eligible candidate on the same representative examples and record task success, factual or formatting errors, tool-use success, latency, and token use. Then compare quality and cost together. You could test a lower-cost model for routine requests and reserve a stronger one for cases that fail your quality checks; that is an implementation choice to evaluate, not an automatic capability promised by the catalog.
API availability is not the same as ChatGPT availability
OpenAI’s API catalog says its latest models are available through the Responses API and Client SDKs. That catalog statement does not, by itself, establish that every API account can use every model or that a model is available to every ChatGPT user. Separately, OpenAI’s ChatGPT help page says GPT-6 is rolling out to eligible users and may not yet be available for a particular account. Check the documentation for the channel you intend to use.
A third-party LifeArchitect page reports that Astra was released to partners on September 3, 2026. The page also labels that date a “public” release while describing the initial launch as partners-only, so its account does not establish general availability. Use OpenAI’s current API or ChatGPT documentation for channel-specific access information.
Production checks before deployment
Before shipping, confirm the current model ID, pricing, limits, supported tools, and access for your API organization in OpenAI’s documentation. The catalog lists Functions, Web search, File search, and Computer use for these four models, but a catalog listing is not a complete setup guide. Consult the current API reference for authentication, request format, tool configuration, and rate limits. (OpenAI API model catalog)
Finally, test the exact workflow you intend to deploy. A large context window does not guarantee that every task will use long context well, and shared listed limits do not mean models have identical quality, speed, or tool behavior. Choose based on measured results and expected token use—not on the word “next” alone.
Sources
GPT-6 Astra Model | OpenAI API