ONLINEAGENT_OPS 2026.Q3 HOME HUBSMODELSBUILDINGCOMPANIESFILMMAKINGSECURITYLIVING WITH ITSKILLSALL HUBS, ONE TERMINAL → ARTICLES CRAFT RECORD BLOG TOOLS SKILLS SEARCH /
HOME›ARTICLES›GPT-6 SOL AND LUNA, EXPLAINED
ARTICLES · EXPLAINERBEGINNER

GPT-6 Sol and Luna, Explained

OpenAI’s GPT-6 family in plain words: what Astra, Sol and Luna are for, what replaced GPT-5.6, which ChatGPT plans get them, and Astra’s Critical cyber rating.

READ8 min
WORDS1,643
SECTIONS6
TYPEEXPLAINER
CHECKED1 OCT 26

GPT-6 is OpenAI’s newest family of models, and it comes in three sizes: Astra, Sol and Luna. Astra is the most capable and was released first, on 3 September 2026. Sol is nearly as good for less money, and Luna is the cheapest, built for simple jobs done in bulk; both arrived on 22 September. A newer Sol, GPT-6.1 Sol, followed on 29 September.1 In ChatGPT, GPT-6 Astra and both Sols are on the paid Plus and Pro plans. Which model the Free and Go plans get is unclear: on 13 Oct 2026 OpenAI’s pricing page lists GPT-6 Luna for them, while its help centre says they get GPT-5.6 Luna.23

TL;DR — THE SHORT VERSION
  • Three models, one family. Astra (strongest), Sol (the balance), Luna (cheapest and fastest).
  • Sol is already on version 6.1. GPT-6.1 Sol came out on 29 September and now takes Sol’s place on OpenAI’s list.
  • They replace GPT-5.6 at the top of OpenAI’s list. GPT-5.6 Sol, Terra and Luna are still sold. Whether GPT-5.6 Luna still runs ChatGPT’s free plan depends on which OpenAI page you read (see below).
  • ChatGPT: Astra and Sol are for paid plans. OpenAI’s plan comparison marks them “No” on Free and Go. On 13 Oct 2026 it marks GPT-6 Luna “Yes” for Free and Go, but OpenAI’s help centre still says those plans get GPT-5.6 Luna.
  • Astra is rated “Critical” for cyber ability — the first OpenAI model at that level. That is a warning label, not a feature.
◈ IN PLAIN TERMS

A model is the AI system that writes the answers; ChatGPT is the app you use to talk to one. OpenAI names its models with a number for the generation (GPT-6 is newer than GPT-5.6) and a name for the tier: Astra, Sol or Luna. Think of it as the same car line in three engine sizes. A bigger model is better at hard problems but slower and costs more; a smaller one is quick and cheap but makes more mistakes on hard work.

What each of the three is for

OpenAI gives each model a one-line job description, and a rule for choosing: start with Astra if you are unsure, pick Sol to balance quality and cost, and Luna for cheap work at large volume.4

OpenAI’s own descriptions, read at source 1 Oct 2026, with a plain reading
ModelOpenAI saysIn everyday words
GPT-6 Astra“Our most capable model for the most demanding work.”The heavyweight. For hard problems where a mistake is expensive: tricky code, long research, careful analysis.
GPT-6.1 Sol“Near-Astra performance for complex work at a lower cost.”The workhorse. Most serious everyday work: coding, documents, long multi-step tasks.
GPT-6 Luna“Our most efficient model for focused, high-volume tasks.”The quick one. Simple jobs you run thousands of times: sorting messages, tagging, short summaries.

OpenAI, Models, read at source 1 Oct 2026. The “in everyday words” column is this site’s reading of OpenAI’s descriptions, not OpenAI’s words.

All three are “reasoning” models: they work through a problem before answering. OpenAI says Sol and Luna take text and images in and give text back.5

This page does not list prices. For what each model costs to use through the API, see OpenAI’s pricing page.

TAKEAWAY

Do not pick by name. Try the cheapest model that might work on a sample of your real tasks, and move up only where it falls short.

Why there is already a GPT-6.1 Sol

A week after GPT-6 Sol, OpenAI released GPT-6.1 Sol. It describes the new version as offering “near-Astra performance for complex work at a lower cost than Astra”, and suggests it for “repeated, long-running work across code, apps, and documents.”6

OpenAI’s models page now features GPT-6.1 Sol and no longer shows the original GPT-6 Sol. The older one has not been switched off: it still appears on the API pricing page and in ChatGPT’s plan comparison, and no retirement date had been published by 1 October 2026.7

The same day, OpenAI also added a faster “Ultrafast” mode for Astra in its API.8

What happened to GPT-5.6

Until September 2026, OpenAI’s featured list was Astra plus three GPT-5.6 models: Sol, Terra and Luna. The GPT-6 models have taken their places. GPT-5.6 Sol gives way to GPT-6.1 Sol, and GPT-5.6 Luna to GPT-6 Luna. Terra, the middle GPT-5.6 model, has no GPT-6 version on the models page.4

“Replaced” here means replaced on the list, not switched off. GPT-5.6 is still sold through the API, and OpenAI’s own deprecation page still names GPT-5.6 models as what to move to from older models it is retiring.9

The full lineup, and how to read OpenAI’s long catalogue, is on which OpenAI model to use.

Which ChatGPT plans get them

On OpenAI’s ChatGPT pricing page, read again on 13 Oct 2026, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Sol are marked “No” for the Free and Go plans. Plus gets all four GPT-6 models, and Pro gets “Expanded” use.

For the free model, OpenAI’s own pages disagree. The pricing page now says Free gets “Unlimited text chats with GPT-6”, and marks GPT-6 Luna “Yes” and GPT-5.6 Luna “No” for Free and Go.2 OpenAI’s help page on ChatGPT’s models, read the same day, says the opposite: “Free and Go users receive GPT-5.6 Luna and do not have access to GPT-5.6 Sol.”3 In this site’s reading, the model menu in your own app is the surest check until the two pages agree.

ChatGPT’s plan comparison, US page, re-read 13 Oct 2026 (OpenAI’s help centre disagrees on Free and Go: see above)
ModelFree and GoPlus / Pro
GPT-6 AstraNoYes / Expanded
GPT-6.1 SolNoYes / Expanded
GPT-6 SolNoYes / Expanded
GPT-6 LunaYesYes / Expanded
GPT-5.6 LunaNoYes / Unlimited*

OpenAI, ChatGPT Pricing, plan comparison, re-read directly at source 13 Oct 2026 (US English page): “Plan: Free, Feature: GPT-6 Luna, Yes”; “Plan: Go, Feature: GPT-6 Luna, Yes”; “Plan: Plus, Feature: GPT-6 Luna, Yes”; “Plan: Pro, Feature: GPT-6 Luna, Expanded”; “Plan: Free, Feature: GPT-6 Astra, No”, and the same No / Yes / Expanded pattern for GPT-6.1 Sol and GPT-6 Sol; “Plan: Free, Feature: GPT-5.6 Luna, No”; “Plan: Pro, Feature: GPT-5.6 Luna, Unlimited*”. The asterisk is OpenAI’s: “Usage must be reasonable and comply with our policies.” The Free plan’s headline reads “Unlimited text chats with GPT-6”, and Plus “Advanced reasoning models with GPT-6”. Read in a browser on 1 Oct 2026, the same page marked GPT-6 Luna “No” and GPT-5.6 Luna “Yes” for Free and Go, and the Free headline read “Unlimited text chats with GPT-5.6 Luna”.

OpenAI adds that for GPT-6.1 Sol, “Availability depends on your plan, client, and workspace settings”, so what you see can differ by app and by country. Business and Enterprise plans were not checked here.10

In Codex, OpenAI’s coding tool, version 0.157.0 of the command-line app added GPT-6 Sol and Luna on 25 September.11

Whether paying is worth it is on free vs paid AI chat.

TAKEAWAY

If you use ChatGPT for free, you get Luna: GPT-6 Luna by OpenAI’s pricing page, GPT-5.6 Luna by its help centre (13 Oct 2026). GPT-6 Astra and Sol are a reason to pay only if the free model is falling short on your work.

Astra’s “Critical” cyber rating, explained

OpenAI rates GPT-6 Astra “Critical” for cybersecurity ability, the top level of its own risk scale, and says it is the first of its models to get there. In plain words: OpenAI believes Astra is good enough at hacking to be dangerous in the wrong hands, so it added extra protections.12

The rating comes from OpenAI’s Preparedness Framework, its own system for measuring whether a model’s abilities could cause severe harm. A system card is the safety report OpenAI publishes with a model. Astra’s says what “Critical” means in practice: given the right tools and access, the model “can find previously unknown security flaws and develop new ways to exploit them across many well-protected systems without a person guiding each step.”13

The same report rates Astra “High”, one step lower, for biological and chemical risk, and below “High” for AI self-improvement (an AI speeding up its own development).13

OpenAI says it responded by strengthening its cyber protections “across the stack” and updating its Trusted Access for Cyber program, which lets vetted security teams use more of the model’s ability for defence. It also tightened security around the model inside OpenAI.13

◈ WHAT IT MEANS FOR YOU

For most people using ChatGPT, the rating is background. It describes what the model could do in skilled hands with the right tools, and why OpenAI put extra protections around it. For companies, it is a reminder that a tool that finds security flaws to fix can also find flaws to exploit. And the rating is OpenAI’s own judgement of its own model: no outside body issued it.

What this page could not verify

  • Any performance claim. “Near-Astra performance” and “most capable” are OpenAI’s words.
  • Whether GPT-6 Luna reaches Free or Go anywhere. A ChatGPT changelog entry read by this site on 29 Sep 2026 pointed to Luna on Free and Go in the desktop app. On 1 Oct 2026 that entry could not be found, and the pricing page marked GPT-6 Luna “No” for both plans. On 13 Oct 2026 the pricing page marks it “Yes” for both, while OpenAI’s help centre still says Free and Go get GPT-5.6 Luna. This page reports both and does not pick one.
  • Plans outside the US page. The plan comparison was read from the US version; other countries and Business or Enterprise plans may differ.
  • How long GPT-6 Sol and GPT-5.6 stay available. Neither had a published shutdown date on 1 Oct 2026.
  • Anything announced at OpenAI’s DevDay on 29 September 2026 beyond the changelog entries quoted above.
SOURCES
  1. 1OpenAI, API changelog, read at source 1 Oct 202622 Sep 2026, “Released GPT-6 Sol and GPT-6 Luna.”; 29 Sep 2026, “Released GPT-6.1 Sol for complex coding and professional work at a lower cost than GPT-6 Astra.” OpenAI, GPT-6 Astra System Card, “Published September 3, 2026”, read in a browser 1 Oct 2026. ↑
  2. 2OpenAI, ChatGPT Pricing, plan comparison, re-read directly at source 13 Oct 2026 (US English page; prices not re-checked): Free “Unlimited text chats with GPT-6”; “Plan: Free, Feature: GPT-6 Luna, Yes”; “Plan: Go, Feature: GPT-6 Luna, Yes”; “Plan: Free, Feature: GPT-6 Sol, No”; “Plan: Free, Feature: GPT-6 Astra, No”; “Plan: Free, Feature: GPT-5.6 Luna, No”.↑a↑b
  3. 3OpenAI Help Center, GPT-6 and other models in ChatGPT, “Updated: 3 days ago”, read at source 13 Oct 2026“Free and Go users receive GPT-5.6 Luna and do not have access to GPT-5.6 Sol.” and “GPT-6 is rolling out in Chat to eligible Pro, Plus, Enterprise, and Business users.” ↑a“Free and Go users receive GPT-5.6 Luna and do not have access to GPT-5.6 Sol.” and “GPT-6 is rolling out in Chat to eligible Pro, Plus, Enterprise, and Business users.” ↑b
  4. 4OpenAI, Models, read at source 1 Oct 2026“If you’re not sure where to start, use GPT-6 Astra, our flagship model for complex reasoning and coding. Choose GPT-6.1 Sol to balance intelligence and cost, or GPT-6 Luna for cost-sensitive, high-volume workloads.” ↑afeatured models GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna; the only GPT-5.6 model shown is “GPT-5.6 Cyber”, under specialised models. ↑b
  5. 5OpenAI, API changelog, 22 Sep 2026, read at source 1 Oct 2026“These reasoning models accept text and image inputs and generate text through the Responses and Chat Completions APIs.” ↑
  6. 6OpenAI, ChatGPT and Codex changelog, 29 Sep 2026, read at source 1 Oct 2026“GPT-6.1 Sol offers near-Astra performance for complex work at a lower cost than Astra. Consider it for repeated, long-running work across code, apps, and documents. Availability depends on your plan, client, and workspace settings.” ↑
  7. 7OpenAI, Models, Pricing and Deprecations, all read at source 1 Oct 2026the models page lists “GPT-6.1 Sol” and not “GPT-6 Sol”; the pricing page lists both gpt-6.1-sol and gpt-6-sol; no GPT-6 deprecation is listed. ↑
  8. 8OpenAI, API changelog, 29 Sep 2026, read at source 1 Oct 2026“Added Ultrafast mode for GPT-6 Astra in the Responses API.” ↑
  9. 9OpenAI, Pricing and Deprecations, both read at source 1 Oct 2026gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna listed with prices; recommended replacements include gpt-5.6-terra, gpt-5.6-sol and gpt-5.6-luna. ↑
  10. 10OpenAI, ChatGPT and Codex changelog, 29 Sep 2026, read at source 1 Oct 2026.↑
  11. 11OpenAI, ChatGPT and Codex changelog, read at source 1 Oct 2026, Codex CLI 0.157.0“Added GPT-6 Sol and Luna, including Amazon Bedrock support and migration prompts for older models.” ↑
  12. 12OpenAI, GPT-6 Astra System Card, 3 Sep 2026, read in a browser 1 Oct 2026“Astra is our first model to reach the Critical level of cybersecurity capability under our Preparedness Framework.” ↑
  13. 13OpenAI, GPT-6 Astra System Card, read in a browser 1 Oct 2026“The Preparedness Framework is OpenAI’s approach to tracking and preparing for frontier capabilities that create new risks of severe harm.” and “This means that, with the right tools and access, GPT-6 Astra can find previously unknown security flaws and develop new ways to exploit them across many well-protected systems without a person guiding each step.” ↑a“we determined that Astra reaches the Critical level in Cybersecurity capability, and the High level in the Biological and Chemical category. In AI Self-Improvement, Astra does not reach our High threshold.” ↑b“we have strengthened our cyber safeguards across the stack, and have updated our Trusted Access for Cyber program, to mitigate the increased risks associated with a cyber critical model.” and “stricter isolation, checkpoint encryption, universal monitoring of full trajectories including chains of thought (CoT), and a blocking alignment evaluation process before internal use.” ↑c

OpenAI developer docs: Models, API changelog, Pricing and Deprecations, read at source on 1 October 2026. OpenAI, ChatGPT and Codex changelog, read at source on 1 October 2026. OpenAI, ChatGPT Pricing, read in a browser on 1 October 2026. OpenAI, GPT-6 Astra System Card, published 3 September 2026, read in a browser on 1 October 2026.

A new model family changes fast in its first weeks. Re-read OpenAI’s pages before you build on any of this.

Free printable AI cheat sheets, and AI news every week.

Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.

SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK
aipromptgeneer://$ about this siteAI PROMPT GENEERAI explained in plain English. Every claim links to its source, with the date it was checked.NEWSLETTERWhat changed in AI each week, in plain English.SUBSCRIBE FREE ↗AI PROMPT GENEER · MACHINE ASSISTANCE DISCLOSEDEVERY ARTICLE CARRIES ITS OWN CHECKED DATE