# GPT-6 Sol and Luna: which one should you actually use?

> Use GPT-6 Sol for coding and complex work, Luna for quick everyday tasks.

*OpenAI just made its two everyday models cheaper and more capable. Here's what Astra, Sol and Luna are each for, what's free, and what the price cut means if you build on any of them.*

By The SuggestedTech Team · SuggestedTech
Canonical: https://suggestedtech.com/news/gpt-6-sol-luna-which-should-you-use

OpenAI has three GPT-6 models now, and until this week most people only knew about one of them. GPT-6 Astra launched three weeks ago as the flagship — the smartest, priciest option. On 22 September 2026, OpenAI filled in the other two: GPT-6 Sol and GPT-6 Luna, both now available, and both considerably cheaper than the versions they replace. If you've been wondering which one you actually want, here's the plain-English version.

## What each model is actually for

Think of it as three tools for three different jobs, not three tiers of the same brain. Astra is for the hardest problems — the ones where getting it right matters more than getting it fast or cheap. Sol is the everyday workhorse: TechCrunch describes it as built for complex tasks like coding, the kind of job where a model needs to hold several steps in mind at once. Luna is for the small, repetitive stuff — the equivalent of asking a very fast, very cheap assistant to summarise a document or pull one fact out of a page, rather than think something through.

> Sol is designed for complex tasks like coding, while OpenAI says Luna is more appropriate for clerical work
> — [TechCrunch](https://techcrunch.com/2026/09/22/openai-launches-gpt-6-sol-and-luna/), 2026-09-22

That framing for Luna is worth keeping in mind, because it tells you exactly when to reach for it: if what you're doing is closer to 'read this and tell me the key point' than 'think this through step by step,' Luna isn't a downgrade — it's the right tool, and it now costs a fraction of what everything else does.

## What it costs now, and what changed

Here's the number that actually matters if you're paying for API access, or using an app built on one of these models: both new models are half the price of the versions they replace, or better. Sol drops from $4 to $2 per million tokens of input and $20 to $10 for output; Luna drops from $0.20 to $0.10 for input and $1.20 to $0.50 for output. A 'million tokens' is roughly 750,000 words, so unless you're building something that talks to the API constantly, this isn't a cost you'll ever notice directly — but if you use ChatGPT, a coding tool that runs on Sol, or any app that quietly calls Luna behind the scenes, you're the reason this price cut exists: OpenAI wants more of that everyday, high-volume use, and cheaper access is how it gets it.

| Model | Best for | Price now (in/out per million tokens) | Price before |
| --- | --- | --- | --- |
| **GPT-6 Astra** | The hardest jobs — deep research, big codebases | $10 / $50 | unchanged |
| **GPT-6 Sol** | Everyday coding and multi-step work | $2 / $10 | $4 / $20 |
| **GPT-6 Luna** | Quick tasks — summarising, extracting, quick answers | $0.10 / $0.50 | $0.20 / $1.20 |

## Who gets what, and what's free

Access depends on what you're using and what you pay for. If you're on ChatGPT Free or Go, you now get Luna in the desktop app — a real upgrade, since a model this capable wasn't previously part of the free tier in this way. Sol needs a paid plan: ChatGPT Work and Codex on Plus, Pro, Business, Enterprise or Edu all get both models the same day. If you use GitHub Copilot for coding, both models arrived there too: Luna on every paid Copilot tier, right down to plain Pro, and Sol from Copilot Pro+ upward. None of this needs you to do anything — OpenAI is rolling it out through the model picker gradually, so if you don't see Sol or Luna yet, it's worth checking back later in the day rather than assuming something's wrong.

One caveat, sourced directly from OpenAI: its own tests show Sol now makes about half as many mistakes as the model it replaces, which is a genuinely useful improvement if you've been burned by a confident wrong answer before. But that number comes from OpenAI's own evaluation of cases where a previous model had already been flagged as wrong — a deliberately hard test, not typical daily use — so treat 'half as many mistakes' as a real trend, not a guarantee for your specific question.

> makes about half as many mistakes as its predecessor, approaching Astra-level reliability at much lower cost
> — [OpenAI](https://openai.com/index/introducing-gpt-6-sol-and-luna/), 2026-09-22

## The practical advice

If you're choosing today: reach for Luna for anything quick and low-stakes, Sol for anything with real steps to it — writing or fixing code, working through a document, planning something with several parts — and only pay for Astra when the job is genuinely hard enough that getting it right the first time matters more than the cost. And if you're building a product on any of these models rather than just using ChatGPT, the price cut is the real story: cheaper Sol and Luna access means whatever you're building just got meaningfully less expensive to run, with no change in how it behaves.

## Key takeaways

- OpenAI released two new everyday models, GPT-6 Sol and GPT-6 Luna, on 22 September 2026, at half the API price of the versions they replace.
- Sol is for complex work — coding, analysis, anything with several steps. Luna is for quick, high-volume jobs — summarising, extracting, answering simple questions.
- Free and Go users can now use Luna through the ChatGPT desktop app; Sol needs a paid plan (Plus, Pro, Business, Enterprise or Edu).
- GitHub Copilot added both models the same day: Sol for Pro+, Max, Business and Enterprise; Luna for every paid Copilot tier including plain Pro.
- GPT-6 Astra, OpenAI's most capable model, didn't get cheaper — it's still the one to reach for on the hardest jobs, at five times Sol's price.

## FAQ

### Which GPT-6 model should I use — Astra, Sol or Luna?
Astra for the hardest jobs, Sol for everyday coding and multi-step work, Luna for quick tasks like summarising or answering a simple question. Sol and Luna both just got 50% cheaper.

### Is GPT-6 Luna free?
Free and Go users can access Luna through the ChatGPT desktop app. Sol needs a paid plan — Plus, Pro, Business, Enterprise or Edu.

### Does GPT-6 Sol work in GitHub Copilot?
Yes, from Copilot Pro+ upward (Pro+, Max, Business, Enterprise). Luna is available on every paid Copilot tier, including plain Pro.

### Is GPT-6 Sol actually more accurate than before?
OpenAI says it makes about half as many mistakes as GPT-5.6 Sol on its internal factuality test — a genuine improvement, though it's OpenAI's own evaluation, not an independent one.

### Did GPT-6 Astra also get cheaper?
No. Astra's price is unchanged. Only Sol and Luna — the two models below the flagship — were cut.

## Sources

- [Introducing GPT-6 Sol and Luna](https://openai.com/index/introducing-gpt-6-sol-and-luna/) — OpenAI, 2026-09-22
- [API Pricing](https://platform.openai.com/docs/pricing) — OpenAI, 2026-09-22
- [OpenAI's GPT-6 Sol and GPT-6 Luna now available](https://github.blog/changelog/2026-09-22-openais-gpt-6-sol-and-gpt-6-luna-now-available/) — GitHub, 2026-09-22
- [OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes](https://techcrunch.com/2026/09/22/openai-launches-gpt-6-sol-and-luna/) — TechCrunch, 2026-09-22
- [OpenAI releases GPT-6 Sol and Luna models, slashing API costs 50% or more](https://venturebeat.com/technology/openai-releases-gpt-6-sol-and-luna-models-slashing-api-costs-50-or-more) — VentureBeat, 2026-09-22
- [Claude Opus 5.5](https://www.anthropic.com/news/claude-opus-5-5) — Anthropic, 2026-09-22
- [Introducing Grok 4.7](https://x.ai/news/grok-4-7) — xAI, 2026-09-21
- [GPT-6 Luna: Release, Intelligence, Performance & Price](https://artificialanalysis.ai/models/releases/gpt-6-luna) — Artificial Analysis, 2026-09-22