As of November 9, 2026, OpenAI recommends GPT-6 Astra for complex reasoning and coding, GPT-6.1 Sol to balance capability and cost, and GPT-6 Luna for cost-sensitive, high-volume work. Treat these as starting points, not a universal ranking: test candidates against your workload. The model IDs and rates below are listed in OpenAI’s model and pricing documentation. (Models; Pricing)
Current GPT API models to compare
| Model | API model ID | Selection guidance stated by OpenAI | Standard input price* | Standard output price* |
|---|---|---|---|---|
| GPT-6 Astra | gpt-6-astra |
OpenAI’s flagship recommendation for complex reasoning and coding | $10 / 1M tokens | $50 / 1M tokens |
| GPT-6.1 Sol | gpt-6.1-sol |
OpenAI recommends it to balance capability and cost | $2 / 1M tokens | $10 / 1M tokens |
| GPT-6.1 Terra | gpt-6.1-terra |
A lower-priced option in this comparison; confirm it meets your workload’s requirements | $1 / 1M tokens | $5 / 1M tokens |
| GPT-6 Luna | gpt-6-luna |
OpenAI recommends it for cost-sensitive, high-volume workloads | $0.10 / 1M tokens | $0.50 / 1M tokens |
*Rates shown are USD per million tokens at the listed standard short-context rates. The materials reviewed here do not establish the request-size threshold that determines whether a request qualifies for short-context pricing, nor how that boundary applies to each model. Check the current pricing page for the applicable rate before estimating spend.
The supplied model-page information also reports a 1.05-million-token context limit and 128,000-token maximum output for the flagship models. Confirm the limit for the exact model ID you plan to use in current model documentation; a context limit is not a recommendation to send prompts of that size. (Models)
Choose by workload, then test
Use OpenAI’s stated positioning to form a shortlist, then evaluate candidates on representative requests:
- Complex reasoning or coding: Start by evaluating GPT-6 Astra, OpenAI’s stated recommendation for these workloads.
- Balancing capability and cost: Consider GPT-6.1 Sol, which OpenAI recommends for this balance. The available information here does not establish that it performs nearly as well as Astra, so compare them on your own tasks rather than assuming equivalent performance.
- Cost-sensitive, high-volume work: Include GPT-6 Luna in your evaluation, especially if its listed token rates fit your budget. Confirm that its results meet your quality requirements before using it in production.
- Considering Terra: Its listed rates are lower than Astra’s and Sol’s. The available information here does not establish a particular capability advantage or workload positioning for Terra, so verify its documented features before selecting it for a capability-specific task.
For any shortlist, run the same representative inputs through the candidate models and assess them against your own acceptance criteria—for example, answer quality, required behavior, latency, and cost per completed task. Keep prompts and settings consistent. This is an evaluation method, not a published benchmark or a claim that one model will perform best for your application.
For requirements involving specific modalities or tools, check the documentation for the exact model and API path. Do not assume that every model supports the same inputs, outputs, or tools based on this comparison.
Estimate token costs before rollout
The calculation below illustrates the listed rates using 1 million input tokens and 100,000 output tokens accumulated across multiple requests. These totals are not one request. The calculations apply only if the individual requests qualify for the displayed short-context rates; the request-level threshold is not established here.
- GPT-6 Astra: (1 × $10) + (0.1 × $50) = $15
- GPT-6.1 Sol: (1 × $2) + (0.1 × $10) = $3
- GPT-6.1 Terra: (1 × $1) + (0.1 × $5) = $1.50
- GPT-6 Luna: (1 × $0.10) + (0.1 × $0.50) = $0.15
This is an illustration using the listed rates, not a forecast of an application’s bill. Actual charges depend on applicable pricing conditions and usage; check the official pricing page for current rates before budgeting.
API usage charges are separate from ChatGPT subscriptions. OpenAI says the API platform and ChatGPT have separate billing systems. (Billing guidance)
Verify the model before production
- Confirm the exact API model ID in OpenAI’s current model documentation; do not rely on an old article or copied example alone. (Models)
- Check the requirements that matter for your workload. Verify the exact model’s documented capabilities, limits, and required features in current documentation before building around them.
- Set the model deliberately in an SDK workflow. OpenAI’s Agents SDK guidance recommends explicit model selection rather than relying on a runtime default. It describes selecting a model per agent, setting a run-level model, or using
OPENAI_DEFAULT_MODELas a process-wide fallback for agents without a model. Consult current language-specific SDK documentation for behavior applicable to your setup. (Agents SDK model guidance) - Recheck pricing and access conditions before deployment and when changing the model or usage pattern. Use the current official pricing and model documentation for the exact choice.
Bottom line: Begin with Astra for demanding reasoning or coding, Sol for OpenAI’s stated capability-and-cost balance, and Luna for cost-sensitive, high-volume workloads. Treat Terra as a lower-priced candidate in this comparison, not as a verified specialist option. Test your shortlist, confirm the specific model’s documented requirements, and calculate costs using the pricing conditions that apply to your requests.