GPT-6 Family Explained: Astra, Sol, Terra, Luna, Cyber

OpenAI's GPT-6 line is four public tiers plus two gated specialists, as of 9 November 2026. Astra is the flagship (launched 3 September), Sol is the coding and professional-work workhorse (GPT-6.1 Sol since 29 September), Terra is the mid-priced computer-use model (26 October), and Luna is the cheap, high-volume model. Cyber and Rosalind are restricted to approved customers.

Key Takeaways

• Four tiers, ordered by API price per million tokens (input/output): Astra $10/$50, Sol $2/$10, Terra $1/$5, Luna $0.10/$0.50, per OpenAI's pricing page.
• The tiers did not launch together. Astra came on 3 September, Sol and Luna on 22 September, GPT-6.1 Sol on 29 September, and GPT-6.1 Terra on 26 October, per OpenAI's API changelog.
• The number after the dot is a refresh of one tier, not a new generation. Only Sol and Terra carry "6.1" so far, while Astra and Luna are still plain "6".
• OpenAI describes Luna as the choice for high-volume, well-defined work such as summarizing and extraction, and Sol as the choice for complex tasks such as coding, per TechCrunch.
• GPT-6 Cyber and GPT-6 Rosalind are not general products. Both sit behind approval programs, and Cyber lists at $20/$100 for trusted-access customers.
• "Pro", "Ultra", "Fast" and "Ultrafast" are modes or service tiers on top of these models, not extra model names. Mixing them up is the most common naming mistake.

The naming guide in plain English

OpenAI's names have three layers, and each answers a different question.

Layer 1, the generation number (GPT-5, GPT-6): how new the underlying training run is. GPT-6 Astra arrived roughly a year after GPT-5, according to 9to5Mac.

Layer 2, the decimal (6, 6.1): a refresh of a specific tier. GPT-6.1 Sol replaced GPT-6 Sol a week after launch. GPT-6.1 Terra replaced GPT-5.6 Terra. There is no GPT-6.1 Astra or GPT-6.1 Luna as of today. The Register reported that OpenAI shelved a planned GPT-6.1 Astra in September; I have only seen that claim in that one report, so treat it as reported, not confirmed.

Layer 3, the tier name (Astra, Sol, Terra, Luna): the size and price bracket. The same Sol, Terra and Luna names were used in the GPT-5.6 family from 9 July 2026, where Sol was the flagship, per TestingCatalog. In GPT-6, Astra moved above Sol. So "Sol" does not mean the same price or rank in GPT-5.6 as it does in GPT-6.

A quick mental model: Astra is the expensive senior reviewer, Sol is the capable generalist, Terra is the fast operator for screens and documents, Luna is the high-volume clerk.

The tiers side by side

Sources: OpenAI API changelog for dates and IDs, OpenAI pricing page for prices, as of 9 November 2026. Prices apply to prompts up to 272K input tokens. Above that, input is 2x and output 1.5x, per the same pricing page.

What each tier is for

Astra: the flagship, and the slow-rollout one

Astra launched on 3 September to a limited set of organizations and widened to Plus, Pro, Business and Enterprise users over the following days, per 9to5Mac. Enterprise admins had to switch it on. OpenAI says Astra meets the "Critical" cybersecurity threshold in its Preparedness Framework, which is why access was staged. My earlier posts cover the marketer's view of Astra and when its Ultrafast mode is worth paying for.

Sol: the default for coding and general professional work

GPT-6 Sol and Luna launched together on 22 September. OpenAI said API prices were half those of the GPT-5.6 Sol and Luna models, crediting caching and inference improvements, per TechCrunch. GPT-6.1 Sol followed on 29 September at the same $2/$10, with a multi-agent beta, per the changelog. I covered the switch decision in GPT-6.1 Sol: what changed and should content teams switch. In the Codex CLI, GPT-6.1 Sol became the default model on 7 October, per Releasebot's Codex changelog.

Terra: the newest tier

OpenAI's launch claims, as reported by BleepingComputer: computer-use tasks more than three times faster than Astra and Sol, and "on par" with Astra on one benchmark at one-tenth the per-token price. OpenAI also says Sol and Astra remain better for the most complex coding. These are vendor figures. An analyst quoted by InfoWorld recommended testing on your own workloads, which I agree with.

Luna: cheap volume, and the model behind Decisions

At $0.10/$0.50, Luna is the model to try first for classification, extraction and routing. It also powers the Decisions API, which went generally available on 22 October, per the changelog. I explained the use cases in OpenAI Decisions API for growth teams. Free and Go ChatGPT users can try Luna in the desktop app, per TechCrunch.

Cyber and Rosalind: gated by approval

GPT-6 Cyber (5 November) is for authorized security testing and goes to approved Daybreak Red customers. GPT-6 Rosalind, released 30 October for life sciences research, is available through a trusted-access program, per the changelog. Neither is something a marketing team can sign up for. I wrote a separate post on who can use GPT-6 Cyber and what Daybreak is.

Modes that are not models

If a vendor, agency or colleague says "we use GPT-6 Pro", ask whether they mean the ChatGPT variant or an API model. They are different products with different billing.

What this means for marketers and founders

My read: for most content and SEO workflows, the practical choice is between Sol and Terra, with Luna as the cheap first pass.

• Test Luna on structured chores this week. Meta descriptions, tag extraction, FAQ classification and redirect-map matching. If the output passes your checklist, you have cut that cost by roughly twenty times versus Sol.
• Keep Sol for drafts a human will publish under a name. It is the tier OpenAI points to for professional work, and its price is low enough that the model bill is rarely the issue.
• Try Terra where screens are involved. Auditing a SERP layout, filling a CMS form, reading screenshots. That is what OpenAI built it for, though I have not tested it and the benchmarks are the vendor's.
• Do not default to Astra. At five times Sol's price, it needs a task where a reviewer's judgment matters more than volume.
• Re-run your tests per tier, not per family. A Luna that passed in September is not a 6.1 model. For a test protocol, see how to evaluate a new model.

What we do not know yet

• Whether a GPT-6.1 Astra or GPT-6.1 Luna will ship. OpenAI has not said, in anything I read.
• Context window sizes. None of the launch coverage I opened states them.
• Whether Sol and Luna were refreshed once or will be again before year end. OpenAI's release gaps have been 26 to 98 days between headline models, by my arithmetic from changelog dates.
• Independent benchmark numbers. Most figures above are OpenAI's own.

FAQ

What is the difference between GPT-6 Astra and GPT-6 Sol?

Astra is the flagship, priced at $10 input and $50 output per million tokens. Sol is a cheaper tier at $2 and $10, positioned for coding and professional work. OpenAI says Sol reaches "Astra-level reliability" on one factuality evaluation, so check whether you need the premium.

What is GPT-6 Luna used for?

Luna is the lowest-priced GPT-6 model at $0.10 input and $0.50 output per million tokens. OpenAI recommends it for high-volume, well-defined tasks such as summarizing documents, extracting information and answering quick questions. It also powers the Decisions API.

Is GPT-6.1 Sol the same as GPT-6 Sol?

No, but the price is the same. GPT-6.1 Sol launched on 29 September 2026, a week after GPT-6 Sol, as an updated model for coding and professional work at $2 and $10 per million tokens. It is the version listed on OpenAI's pricing page.

What does Terra mean in the GPT-6 family?

Terra is the mid-tier model, priced at $1 input and $5 output per million tokens. GPT-6.1 Terra launched on 26 October 2026 and targets visual reasoning, computer use and writing. It replaced GPT-5.6 Terra.

Can I use GPT-6 Cyber?

Only if OpenAI approves you. GPT-6 Cyber was released on 5 November 2026 to select approved Daybreak Red customers through the API and Codex. OpenAI said access would widen to all Daybreak Red customers in the coming weeks.

Which GPT-6 model is cheapest?

Luna, at $0.10 input and $0.50 output per million tokens as of 9 November 2026. Batch and Flex processing list at 50% of standard rates for GPT-6 models, per OpenAI's pricing page.

Do Free ChatGPT users get GPT-6?

Free and Go users can try GPT-6 Luna in the desktop app, according to TechCrunch. Astra, Sol and Terra are listed for paid plans in the launch coverage I read. Plan rules change often, so check ChatGPT's current plan page.

Want help with this?

I'm Younus Fardeen, an organic growth, SEO and AEO strategist with 4+ years of marketing experience across edtech and startup teams. If you want help choosing and testing AI models for your content and growth workflows, see my work and reach me through the contact form at younusfardeen.com.