OpenAI’s API models documentation lists four flagship models to start with: GPT-6 Astra (gpt-6-astra) for demanding reasoning and coding, GPT-6.1 Sol (gpt-6.1-sol) for a balance of capability and cost, GPT-6.1 Terra (gpt-6.1-terra) for visual reasoning, computer use, and writing, and GPT-6 Luna (gpt-6-luna) for focused, high-volume work. The page also names specialized models for images, audio, transcription, life sciences, and cybersecurity. It points to a broader catalog, so the models below are a useful starting list—not an exhaustive inventory. OpenAI API models documentation
OpenAI flagship models at a glance
| Model | API model ID | Starting point described by OpenAI | Listed input price per 1M tokens | Listed output price per 1M tokens |
|---|---|---|---|---|
| GPT-6 Astra | gpt-6-astra |
Complex reasoning, coding, and demanding work | $10 | $50 |
| GPT-6.1 Sol | gpt-6.1-sol |
Balancing intelligence and cost on complex work | $2 | $10 |
| GPT-6.1 Terra | gpt-6.1-terra |
Visual reasoning, precise computer use, and natural writing | $1 | $5 |
| GPT-6 Luna | gpt-6-luna |
Focused, cost-sensitive, high-volume tasks | $0.10 | $0.50 |
The OpenAI page lists a 1.05 million-token context window and 128,000-token maximum output for each of these four models. It also lists Functions, Web search, File search, and Computer use under each model’s tools. These are separate catalog fields: the listing does not establish that the maximum context and maximum output can always be used together.
The table reproduces the per-input-token and per-output-token rates displayed in the model catalog. The supplied catalog text does not specify additional pricing conditions such as cached-input treatment, service-tier rates, or input-length thresholds, so these figures alone do not determine a request’s total cost. Confirm the current pricing basis for your intended API usage before estimating production spend. The listing also does not establish account-specific access or lifecycle status for each model; check those separately before deployment.
Which model should you shortlist?
- Start with Astra if the task is demanding reasoning or coding and you want to evaluate OpenAI’s flagship recommendation. OpenAI describes it as its most capable model for demanding work.
- Evaluate Sol if you need complex-work capability at lower displayed token rates than Astra. OpenAI describes its performance as near-Astra at lower cost; that is the provider’s characterization, not an independent benchmark result. The lower displayed rates do not by themselves establish a lower total bill for every workload.
- Consider Terra when visual reasoning, computer use, or clear natural writing is central to the task. The catalog lists Computer use for Terra, but the listing alone does not specify the API and tool configuration required for your intended computer-use workflow; confirm those details in the relevant current API documentation.
- Consider Luna for focused tasks at high volume when cost is a priority. Its displayed token rates are the lowest among these four models, but application-specific accuracy, latency, and output quality still need evaluation.
For example, a team building a coding assistant might put Astra and Sol in its initial comparison; a high-volume workflow that classifies or summarizes short inputs might also test Luna. Those are starting candidates based on OpenAI’s descriptions—not guarantees about results in a particular product. Evaluate shortlisted models using representative inputs, expected output lengths, and your own quality and latency requirements.
OpenAI’s documentation says its latest models support text and image input, text output, multilingual capabilities, and vision, and are available through the Responses API and Client SDKs. Treat that as general orientation: confirm the exact interface and capability for the particular model and feature you plan to use.
Specialized OpenAI models by task
The same API models page names specialist families alongside its flagship models. Examples shown in the documentation include:
- Image generation and editing: GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare, as well as GPT-Image-2.5 Sunburst and GPT-Image-2.5 Flare. The page distinguishes detailed visual work and fast iteration among its descriptions.
- Realtime speech and translation: GPT-Live 1, GPT-Realtime-2.1, GPT-Realtime-2.1 Mini, GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-1.5. The page describes GPT-Realtime-Translate as a streaming speech-to-speech translation model.
- Transcription: GPT-Transcribe, GPT-Live-Transcribe, and GPT-Realtime-Whisper. OpenAI describes GPT-Transcribe for file and Realtime input transcription, and GPT-Live-Transcribe for low-latency speech-to-text.
- Life sciences: GPT-Rosalind, described as a life-sciences reasoning model for approved organizations.
- Cybersecurity: GPT-6-Cyber and GPT-5.6 Cyber, described for defensive or authorized vulnerability research and security testing.
These are examples visible on the documentation page, not a complete specialist catalog. Its descriptions do not provide a comparable set of prices, limits, or detailed API specifications for every specialist model. Check individual model documentation before choosing one for production.
Production checks before you choose
Before building around a model, verify its exact model ID, current access and lifecycle status, and support for the API, modality, and tools your application will call. For computer use with Terra, check the model-specific API and tool requirements rather than relying only on the catalog’s tool label. Confirm current pricing and limits, then test the model against representative inputs and outputs. The catalog descriptions alone do not establish application-specific quality or response time.
The model page identifies the Responses API and Client SDKs as access routes, but a model listing is not a complete implementation guide. Use the relevant current API documentation for request details and model-specific support. OpenAI API models documentation
Direct OpenAI API models are not the same as other catalogs
Availability depends on the provider and channel. Microsoft Foundry’s Azure OpenAI catalog describes models hosted and billed through Azure, and Microsoft says availability there varies by region and cloud. It is useful for Azure deployment decisions, but it does not establish the complete direct OpenAI API catalog. Microsoft Foundry model documentation
Similarly, a third-party software package may list only the models it supports. For instance, the LLM documentation describes its own integration’s supported-model list; that is not a definitive OpenAI catalog. LLM’s OpenAI integration documentation
For direct API selection, use OpenAI’s model documentation for model IDs and capabilities, then confirm current availability and terms in the relevant OpenAI documentation. Do not assume that an Azure deployment list or a third-party integration list applies unchanged to your OpenAI API account.