Select an OpenAI API model, understand its charge categories and connect it to a reproducible request. This hub links the current developer documentation to practical implementation choices.
What OpenAI offers through the API
OpenAI provides the Responses API for model calls through its official SDKs. The developer catalog covers general text and reasoning models as well as specialized capabilities such as embeddings, audio and image generation. Choose the endpoint and model together: the response contract is part of the integration, not just a transport detail. Official documentation.
A practical starting point is a narrow text task whose result you can judge without another model. Ask for a classification from an allowed set, a short extraction from supplied material or an explanation that a reviewer can verify. Keep the first request free of external tools so you can separate basic authentication and model access from the behavior of a larger workflow.
The current catalog names Astra, Sol, Terra and Luna among its model choices and documents API identifiers and capabilities. It also links to specialized model surfaces. These labels help organize a shortlist; they do not tell you which output will meet your application’s acceptance criteria on every input. Official documentation.
For a support assistant, save examples where the correct outcome is escalation rather than a confident answer. For code generation, run the resulting change against the actual project checks. For extraction, include incomplete documents and contradictory fields. Your evaluation should reward the result your product needs, including a useful refusal to guess, rather than a response that simply reads fluently.
Design the application boundary before adding hosted tools. Decide whether a response is only displayed, used to populate a draft, or allowed to request an external action. Record the action and its result independently of the text that accompanies it. This makes a failed step understandable and gives you somewhere to place application-specific review rules. Keep credentials out of user-visible requests and logs.
OpenAI’s production guidance organizes access around projects and provides separate operational controls for spending and request capacity. An experiment and a deployed service should have an identifiable owner and a deliberate configuration. Review those controls in the project that will actually run the workload, rather than assume a successful local call proves production readiness. Official documentation.
Use this hub to connect those decisions. The model list answers what can be selected; Pricing answers how request categories are charged; Rate limits answers how you should pace work. The API-key guide and Python tutorial turn the choice into a small reproducible call. Errors explains which details to preserve when the call fails. Return to the same evaluation set when changing a model or endpoint so improvements and regressions stay visible.
Before expanding the prototype, write a short operating note covering the selected identifier, the request shape, the secret’s owner and the acceptable failure state. Include where to find the provider documentation and how to stop the application’s scheduled work. This note is more useful than a copied example that nobody knows how to maintain.
Pricing at a glance
| Summary | Verified value |
|---|---|
| Lowest input price model | gpt-5.6-luna |
| Input price | $0.2 per 1M tokens |
| Output price | $1.2 per 1M tokens |
| Free access | No universal ongoing free inference allowance was established in the last successful check. |
Last verified · Source ↗
Review OpenAI API pricing before projecting a workload.
Models available
| Model | Official identifier | Modalities | Context tokens | Input USD / 1M | Output USD / 1M | Status |
|---|---|---|---|---|---|---|
| gpt-5.6-luna | gpt-5.6-luna | text | 1,050,000 | $0.2 | $1.2 | Active |
| gpt-5.6-sol | gpt-5.6-sol | text | 1,050,000 | $4 | $20 | Active |
| gpt-5.6-terra | gpt-5.6-terra | text | 1,050,000 | $2 | $12 | Active |
| gpt-6-astra | gpt-6-astra | text | 1,050,000 | $10 | $50 | Active |
Last verified · Source ↗
Use the OpenAI model guide to distinguish identifiers and capabilities.
Free tier and how to get a key
The official billing documentation was checked. No universal ongoing free inference allowance was established by that source. Promotional or account-specific credits require a check in your account. Read the official terms ↗
Last verified · Source ↗
Follow project key setup and verify any account-specific offer.
Rate limits and tiers
Read project and organization rate limits before introducing parallel requests.
Service status
Start in 5 minutes
Follow the Responses API Python tutorial.
result = client.responses.create(model="gpt-5.6-luna", input="Hello", store=False)
print(result.output_text)Recent changes
- OpenAI · gpt-live-1 — Amount UsdNot previously recorded → Price category: per_minute · USD: 0.05 · Unit: per minuteSource ↗
- OpenAI · gpt-live-1 — MetadataModalities: ["text"] → Modalities: ["audio-in","audio-out"]Source ↗
- OpenAI · babbage-002 — Amount UsdNot previously recorded → Price category: batch_output · USD: 0.2 · Unit: per 1M tokensSource ↗
- OpenAI · babbage-002 — Amount UsdNot previously recorded → Price category: batch_input · USD: 0.2 · Unit: per 1M tokensSource ↗
- OpenAI · babbage-002 — Amount UsdNot previously recorded → Price category: output · USD: 0.4 · Unit: per 1M tokensSource ↗
- OpenAI · babbage-002 — Amount UsdNot previously recorded → Price category: input · USD: 0.4 · Unit: per 1M tokensSource ↗
- OpenAI · davinci-002 — Amount UsdNot previously recorded → Price category: batch_output · USD: 1 · Unit: per 1M tokensSource ↗
- OpenAI · davinci-002 — Amount UsdNot previously recorded → Price category: batch_input · USD: 1 · Unit: per 1M tokensSource ↗
Use the AI API cost calculator to turn the model and workload you are considering into an estimate.
Last verified · Source ↗
Frequently asked questions
Which interface does this starter use?
Can I substitute a model display name?
Where should a production key live?
Does every capacity error mean I should retry?
How do I compare a model change?
Sources
- OpenAI developer quickstart ↗
- OpenAI models ↗
- API pricing ↗
- Rate limits ↗
- Error codes ↗
- Production best practices ↗
- Batch API ↗
- Spend limits ↗
Last verified · Source ↗