OpenAI GPT-6 family
OpenAI Models and Token Costs: GPT-6, 6.1 and 6.2
Which OpenAI models are current, how the Astra, Sol, Terra and Luna tiers differ, what each costs per million tokens, and how to choose between them for production work.
Prices and versions: figures are published list prices in US dollars per 1 million tokens, read from each provider’s own pages on 18 November 2026. Models, prices and promotions change often, so confirm on the provider’s page before you budget.
The current model line
OpenAI’s current text models come in four named tiers. Astra is the top tier, Sol is the balanced workhorse, Terra is fast and efficient, and Luna is the lowest-priced, most efficient tier. Numbered releases refresh a tier: GPT-6.1 updated Sol and Terra, and GPT-6.2 added newer Astra and Sol versions. Older generations stay listed in places, some with promotional pricing or deprecation notices, so pin the exact model id rather than a tier name.
Token prices by model
List prices in US dollars per 1 million tokens at the standard processing tier. Cached input is the discounted rate for repeated prompt prefixes.
| Model | API id | Input | Cached input | Output | Context | Positioning |
|---|---|---|---|---|---|---|
| GPT-6.2 Astra | gpt-6.2-astra | $8.00 | $0.40 | $40.00 | Not on page | Top tier, newest Astra |
| GPT-6.2 Sol | gpt-6.2-sol | $2.00 | $0.10 | $10.00 | Not on page | Balanced, newest Sol |
| GPT-6 Astra | gpt-6-astra | $10.00 | $1.00 | $50.00 | 1.05M | Flagship, most capable |
| GPT-6.1 Sol | gpt-6.1-sol | $2.00 | $0.10 | $10.00 | 1.05M | Balanced |
| GPT-6.1 Terra | gpt-6.1-terra | $1.00 | Check page | $5.00 | 1.05M | Fast and efficient |
| GPT-6 Luna | gpt-6-luna | $0.10 | Check page | $0.50 | 1.05M | Most efficient, lowest price |
The models page lists a 128K maximum output for the four models it details. Knowledge cutoffs shown there are 30 April 2026 for Astra and Sol, 1 June 2026 for Terra and 18 May 2026 for Luna. Models released after those pages were written may not show every column.
What a monthly bill looks like
A worked example: an application that sends 100 million input tokens and 20 million output tokens a month, with no caching or batching. Your own mix of input, output and cache hits will move these numbers.
| Model | Monthly cost: 100M input + 20M output tokens |
|---|---|
| GPT-6.2 Astra | $1,600 |
| GPT-6 Astra | $2,000 |
| GPT-6.2 Sol or GPT-6.1 Sol | $400 |
| GPT-6.1 Terra | $200 |
| GPT-6 Luna | $20 |
What changes the bill besides the model
- Processing tier. Standard, Batch, Flex, Fast and Ultrafast are listed. Batch and Flex trade latency for a lower price; Fast and Ultrafast charge a multiple of standard for lower latency.
- Prompt caching. Repeated prefixes bill at the cached-input rate, a small fraction of normal input. Stable system prompts and tool definitions are the usual winners.
- Long prompts. Some models carry higher rates for very long prompts, so check the model row before relying on the full context window.
- Regional endpoints. Data-residency and regulated-cloud endpoints can carry an uplift on newer models.
- Output tokens. Output costs several times input on every tier, and reasoning effort adds tokens you pay for.
Specialised and open-weight models
OpenAI also lists realtime voice models, restricted-access models for areas such as agent security and defence and life sciences, and legal-domain configurations, each with its own price. Separately, OpenAI publishes open-weight models (gpt-oss-120b and gpt-oss-20b) that third-party hosts serve at very low prices; see Hugging Face models for current hosted rates and open-source AI models for when to self-host them.
How to choose a tier
- Start with Sol or Terra. Most production assistants, extraction and coding agents land here. Move up only when your own evaluations fail.
- Use Luna for volume. Classification, routing, tagging and first-pass triage at the lowest price.
- Reserve Astra for hard cases. Long-horizon reasoning and difficult agentic work, ideally behind a router that escalates only when needed.
- Measure on your data. Public benchmarks do not predict your cost per solved task. Run a small evaluation set and compare cost per success, not price per token.
Where this meets infrastructure
Token price is only one side of the decision. High-volume or sensitive workloads can be cheaper or more compliant on hosted or self-run open models, and agents multiply token use. We help teams compare API, hosted models and self-hosted capacity, and wire the result into the agent platform and the application layer. Unit economics also depend on the GPUs behind the endpoint.
Where this fits in the stack
Common questions
Which OpenAI model is cheapest?
On the pages we read, GPT-6 Luna is the lowest-priced current text model at $0.10 per million input tokens and $0.50 per million output tokens. Older or open-weight models can be cheaper still on third-party hosts.
What is the difference between Astra, Sol, Terra and Luna?
They are performance and price tiers: Astra is the most capable, Sol is balanced, Terra is fast and efficient, and Luna is the most efficient and lowest priced. Version numbers such as 6.1 and 6.2 refresh a tier.
How do I reduce OpenAI API costs?
Pick the smallest tier that passes your evaluations, cache stable prompt prefixes, use Batch or Flex for work that can wait, cap output length, and route only hard cases to the top tier.
Are these prices final?
No. They are list prices read on 18 November 2026. Promotions, new releases and deprecations change them, so check the provider pricing page before budgeting.
Sources and dating
Source: OpenAI developer documentation (pricing and models pages), read 18 November 2026. Some page sections were large, so a few cells (cached input for Terra and Luna, context for the 6.2 models) are marked for checking rather than filled in. Some figures are vendor claims or preliminary and some differ between sources, so we confirm against the provider pages before we size or budget a build. OpenAI, GPT and the model names are trademarks of their owners. Embedica is independent and is not affiliated with or endorsed by any of the companies named.
Next steps
Related: LLM models and token costs · Anthropic Claude models · Open-source AI models · AI application layer
Need help choosing between API, hosted and self-run models?
Model and cost review
