Since October, ChatGPT just says “GPT-6” and moves on. Under the hood, there is no such thing as one GPT-6: there’s a whole sky. OpenAI named its range after it: Sol the sun, Luna the moon, Astra the stars. Once you see the sky for what it is (the everyday star, the cheap nearby rock, and the expensive distant lights), choosing a GPT model stops being mysterious. Let’s look up: it’s not rocket science.

A sky, not a single star

Like every vendor, OpenAI sells a model family: one brand, one API, several sizes. The naming does half the explaining. Sol is the sun: the default everything orbits, bright enough for almost any day. Luna is the moon: small, close, incredibly cheap to visit, perfect for quick trips. Astra is the stars: the frontier, spectacular and costly to reach, and you rarely need them to light the kitchen.

ChatGPT quietly assigns you a spot in this sky: since early October, free users chat with Luna, paid tiers get Sol, and Astra is reserved for the Pro plans. You often don’t pick the model; the subscription picks it for you.

The current sky (autumn 2026)

API prices per million tokens, input/output:

Body Model Released Price (in/out) The job
The stars GPT-6 Astra Sep 3, 2026 $10 / $50 Deep reasoning, the hardest problems
The sun GPT-6 Sol Sep 22, 2026 $2 / $10 The everyday default
The sun, retuned GPT-6.1 Sol Sep 29, 2026 $2 / $10 Agents and coding, cached input at $0.10
The moon GPT-6 Luna Sep 22, 2026 $0.10 / $0.50 High volume, drafts, classification

The whole sky shares a 1.05 million-token context window (922K in, 128K out). Luna at $0.10/$0.50 sits at exactly the same price point as Anthropic’s Haiku 5.5: the cheap tier is where the price war is loudest this autumn.

One oddity worth knowing: there is no GPT-6 Terra. The mid-tier name from the GPT-5.6 generation simply didn’t make the trip. And the previous sky still shines: GPT-5.6 Sol, Terra and Luna (July 2026) are still served, like starlight still arriving after the star moved on. Just mind their age: GPT-5.6 Sol’s knowledge cutoff is February 16, 2026; ask it about anything newer and it’s honestly in the dark.

The 272K line: where the meter doubles

A detail that bites real projects. Past 272K input tokens, OpenAI bills the whole request at 2x input and 1.5x output. Send a 400K-token codebase to GPT-6 Sol with a 2K answer:

Naive estimate   400K × $2/M + 2K × $10/M  ≈ $0.82
Actual billing   400K × $4/M + 2K × $15/M  ≈ $1.63

Cross the line by one token and the trip costs double. The 1M-token window is real, but the last 750K tokens of it are premium seats. Trimming context isn’t just hygiene; here it’s directly money.

The numbers (with the asterisks)

House policy, so here they are. Prices first: OpenAI’s pricing page blocks crawlers, so the GPT-6 figures above come from third-party aggregators (OpenRouter and friends). Several agree with each other, but they are not the primary source. Claims second: “GPT-6.1 Sol delivers near-Astra intelligence at a fifth of the price” is OpenAI’s own benchmark, vendor-reported like most launch charts (see our benchmarks article). Safety third: Astra is the first OpenAI model to hit the company’s internal “Critical” cybersecurity threshold, and ships with some cyber capabilities restricted; a planned GPT-6.1 Astra was reportedly scrapped over safety concerns, per the WSJ. The frontier tier, here as at Anthropic, no longer ships unfiltered.

Also in the sky: a Luna Decisions beta (October 6) that returns calibrated probabilities instead of text, OpenAI’s entry in the decision model family we covered in October.

In summary

Question Answer
What is the GPT-6 family? Three bodies: Astra (stars, frontier), Sol (sun, default), Luna (moon, cheap)
Who gets what in ChatGPT? Free → Luna, paid → Sol, Pro plans → Astra
The price range? From $0.10/$0.50 (Luna) to $10/$50 (Astra), per million tokens
Context window? 1.05M tokens everywhere, but 2x billing past 272K input
Is there a GPT-6 Terra? No; Terra only exists in the still-served GPT-5.6 generation
The caveats GPT-6 API prices are aggregator-sourced; launch benchmarks vendor-reported

Next time someone “uses GPT-6”, you can ask which part of the sky: the moon for the mail sorting, the sun for the daily work, the stars when the problem truly deserves an expedition. Look at the price per million, mind the 272K line, and keep your telescope (your evals) pointed at your own tasks: it’s not rocket science.

Sources: Wikipedia, GPT-6 · CNBC, Astra launch and cyber restrictions · OpenAI, introducing GPT-6.1 Sol · TechCrunch, GPT-6.1 Sol and the scrapped Astra · OpenAI, GPT-5.6 Sol model docs · stob.ai, OpenAI pricing comparison · Vercel, Decisions API on AI Gateway.