GPT-6 Sol and Luna Explained: Which Model Your ChatGPT Plan Actually Uses Now, and Whether Claude Opus 5.5 Changes the Math

What's in this guide

• The short version
• Sol, Luna, Terra, Astra: what the names actually mean
• Which model is answering you right now, plan by plan
• Why GPT-6 Sol is not in your model picker
• The same day, Anthropic shipped Claude Opus 5.5
• I ran Sol and Luna on the same three jobs
• Should you pay more now? A decision chart
• What to watch over the next few weeks
• The takeaway

On September 22, OpenAI released two new models called GPT-6 Sol and GPT-6 Luna. About ninety minutes earlier, Anthropic released Claude Opus 5.5. Both made the same pitch: flagship-level results for a lot less money. If you pay for ChatGPT or Claude, that matters, because for the first time in a while the question is not "should I upgrade" but "can I safely stay on the cheap plan".

I spent the week since launch switching between the models on my own Plus account. This is what I wish someone had handed me on day one: what the names mean, which model your plan really gives you (not always the one you think), and a decision chart for the upgrade question.

The short version

GPT-6 Sol is the mid-size model for real work. GPT-6 Luna is the small, fast one. GPT-6 Astra, which came out on September 3, is still the flagship, and OpenAI says plainly that Astra "continues to be our best model across the board."

Sol and Luna are available now to Plus, Pro, Business, Enterprise and Edu subscribers, but only inside ChatGPT Work and Codex. They are not yet in the regular Chat window. Free and Go users can try GPT-6 Luna in the desktop app. Through the API both models cost exactly half of what their GPT-5.6 versions did.

Claude Opus 5.5 is included in the $20 Claude Pro plan at no extra charge, a change from the Fable 5.1 launch, where Pro users had to pay for extra usage.

If you want to skip the rest: stay on ChatGPT Plus or Claude Pro. Both $20 plans got noticeably better this week without a price change. The $100 and $200 tiers are now about usage volume, not a smarter model.

Sol, Luna, Terra, Astra: what the names actually mean

OpenAI's naming used to be a mess of numbers and letters. Since GPT-5.6 it has used celestial names, and once you see the pattern it is tidy.

Luna (the moon) is the small model, built for speed and volume: quick answers, summaries, pulling facts out of documents. It is what Free and Go users have been talking to for months in its GPT-5.6 form.

Sol (the sun) is the mid-size workhorse that most paying users have been using in Chat as GPT-5.6 Sol. It handles coding, multi-step reasoning and longer documents, and it is the one OpenAI pushes hardest for "everyday work."

Astra (the star) is the flagship, the most capable and by far the most expensive to run. In the Chat window it appears under the label GPT-6 Pro, which confused a lot of people at launch. My GPT-6 Astra guide covers who gets it.

Terra (the earth) is the odd one out. It shows up in Codex as GPT-5.6 Terra for Free and Go accounts and is tuned for coding. There is no GPT-6 Terra as of this writing.

So the ladder, from cheapest to most capable, is Luna, Sol, Astra. This release moved the two lower rungs up to the GPT-6 generation.

The price gap is clearest in the API, which is what developers pay per million tokens (roughly 750,000 words). GPT-6 Sol costs $2 to read a million tokens and $10 to write them, half of what GPT-5.6 Sol cost. GPT-6 Luna is $0.10 and $0.50. Claude Opus 5.5 is $4 and $20, and Claude Fable 5.1 is $10 and $50. You will never see those numbers on a ChatGPT bill, but they explain the strategy: Luna is so cheap that OpenAI can afford to hand it to free users.

Which model is answering you right now, plan by plan

ChatGPT is really three products that share one login: Chat (the normal conversation window), Work (the agent that builds documents, spreadsheets and reports for you), and Codex (the coding agent that lives in the desktop app). Each one has its own model list, and the new models landed in only two of the three.

Here is what I see, matched against OpenAI's help pages:

Free and Go ($8 a month). Chat still runs GPT-5.6 Luna. The new GPT-6 Luna is available in the desktop app only, which in practice means you need to download the app and use it there rather than in a browser tab.

Plus ($20 a month). Chat still runs GPT-5.6 Sol with the Instant, Medium and High thinking levels. Open Work or Codex and the picker changes: on my Plus account it lists GPT-6 Luna High, GPT-6 Sol Light, GPT-6 Sol Medium, and a limited amount of GPT-6 Astra at Light and Medium.

Pro ($100 and $200 a month). Chat runs GPT-5.6 Sol plus GPT-6 Pro, which is Astra under a different label, capped at 50 messages a week on the $100 plan and 200 a week on the $200 plan. Work and Codex get Sol, Luna and Astra with the higher allowances.

Worth knowing: Work and Codex share a usage allowance that is separate from Chat, so a heavy afternoon in Work will not eat into your Chat messages.

Why GPT-6 Sol is not in your model picker

If you opened the model dropdown this week and saw nothing new, that is expected. OpenAI's own announcement says the models "are not yet available in Chat." It is a rollout choice, not a bug, and it is the same pattern Astra followed earlier this month.

The workaround is simple. On the left sidebar (or the bottom tab bar on mobile) there is a Work entry. Open it, start a task, and the model picker inside shows the GPT-6 options. Work is built for jobs you hand off rather than conversations you steer, but nothing stops you from asking it a plain question to see how Sol answers. Codex is the same idea for code and lives only in the desktop app, so non-programmers can ignore it.

Why keep them out of Chat? OpenAI has not said, but Chat carries most of ChatGPT's traffic, and capacity is the obvious reason. Expect Chat to follow in weeks.

The same day, Anthropic shipped Claude Opus 5.5

Opus 5.5 is the new top of Anthropic's regular three-model lineup (Opus, Sonnet, Haiku). Anthropic says it performs at the level of Claude Fable 5.1 on most work and costs 40 percent less to run than Opus 5. It is also more direct: Anthropic says it puts the important information first and cuts the padding that made Opus 5 tiring to read. If you read my Claude Fable 5.1 guide, the headline is that most people no longer need Fable for everyday tasks.

Three consumer-facing changes matter more than the benchmarks:

1. Opus 5.5 is included with Claude Pro ($20). Pro, Max and Team subscribers get it at no extra cost. Free users stay on Sonnet and Haiku.
2. The five-hour usage limits went up on Pro, Max, Team and seat-based Enterprise plans. Anthropic has not published the new numbers, but the model itself is cheaper to run, so each session goes further.
3. Every subscriber got a saved rate-limit reset that you can spend whenever you hit the wall. According to Notebookcheck, the reset in the app expires on October 22, so use it before then.

One odd detail from the launch post: questions on certain cybersecurity or biology topics are quietly handed to an older model (Opus 4.8 or Opus 5) as a safety measure. For ordinary use you will never notice. Sonnet 5.5 and Haiku 5.5 will "follow in the coming weeks."

I ran Sol and Luna on the same three jobs

Benchmarks are OpenAI's numbers on OpenAI's tests. I wanted to know what the difference feels like, so I gave GPT-6 Luna, GPT-6 Sol and GPT-5.6 Sol (the model Plus users still get in Chat) the same three tasks through Work on my Plus account.

Job one: summarise a 38-page rental contract into a one-page list of what I can and cannot do. Luna finished in about 20 seconds and got the list right but missed the sublet clause on page 31. Sol took closer to a minute, caught it, and added a note that two clauses contradicted each other, which they did. GPT-5.6 Sol also caught the clause but used twice as many words to say so.

Job two: clean up an expenses spreadsheet with mixed date formats and duplicate rows. All three managed it. Luna was fastest by a wide margin. Sol was the only one that flagged three rows where the amount looked like a typo (a decimal point in the wrong place). That is the "fewer mistakes" claim showing up in practice.

Job three: draft a complaint email to an airline in a firm but polite tone. This is where Luna shines. The email was short, sounded like a person, and needed no edits. Sol's was also good but a paragraph too long. GPT-5.6 Sol's version had the over-hedged, jargon-heavy feel OpenAI says it has trained out of the new models, and in my test that claim holds up.

My takeaway after a week: Luna is the right default for anything you would previously have typed into Chat. Sol is worth the wait for anything with money, contracts or numbers in it. I never once wished I had Astra.

Should you pay more now? A decision chart

You are on ChatGPT Free. Download the desktop app and try GPT-6 Luna. It is a bigger upgrade than moving to Go ($8), which still runs the older GPT-5.6 Luna in Chat. Pay for Go only if you need more uploads and image generation.

You are on ChatGPT Plus. Stay. You now get GPT-6 Sol and Luna in Work and Codex, a taste of Astra, and Chat will follow. Pro ($100) buys you 50 GPT-6 Pro messages a week in Chat, which is a lot of money for a model that OpenAI's own charts put only a little ahead of Sol on everyday work.

You are on ChatGPT Pro. The $200 tier makes sense only if you hit the 50-a-week cap on the $100 tier or run Work and Codex all day. If you upgraded for Astra's smarts rather than for volume, Sol now covers most of that ground and you could drop back to Plus.

You are on Claude Pro. Stay, and enjoy Opus 5.5 at no extra cost. Max ($100 or $200, for 5x or 20x the usage) is for people who run Claude Code or long agent sessions for hours. The raised five-hour limits make Pro stretch further than it did last week.

You are choosing between them. For long documents and writing quality I still lean Claude. For spreadsheets, Codex and the widest range of built-in tools I lean ChatGPT. My Claude vs ChatGPT vs Gemini comparison goes deeper. What changed this week is that both $20 plans are now clearly the value pick.

What to watch over the next few weeks

GPT-6 Sol and Luna in Chat. OpenAI's wording is "not yet," which usually means weeks. When it happens, Plus users get a real upgrade in the window they use most.

Sonnet 5.5 and Haiku 5.5. Haiku 5.5 is the one to watch for Claude Free users, since Free accounts run Sonnet and Haiku.

Gemini 4. Google told 9to5Google on September 24 that Gemini 4 is coming "as soon as possible." No date, and no word on which plans get it. My Gemini 3.8 Live guide covers what Google AI subscribers have today.

Your usage limits. Both companies moved the goalposts this week. Check what your plan actually allows, and if you are on Claude, spend that saved reset before October 22.

The takeaway

The September 22 launches were not about a smarter AI. They were about the smart AI getting cheap enough to hand to everyone. GPT-6 Sol gives Plus users most of what Astra offered three weeks ago, Luna gives free users a model that would have been a paid feature a year ago, and Claude Opus 5.5 gives Claude Pro subscribers a model that beats Fable on several of Anthropic's own tests.

If you pay $20 a month to either company, you got a free upgrade this week. If you pay $100 or more, take a hard look at whether you are buying intelligence or just volume, because as of this week the intelligence is on the cheap plan.