OpenAI just gave its GPT-6 family two more members, and the headline is price, not power. On September 22, 2026, OpenAI introduced GPT-6 Sol and GPT-6 Luna, filling out the mid-tier and budget slots beneath GPT-6 Astra, the flagship model that launched September 3, 2026 (LumiChats already covered how Astra stacks up against Fable 5.1 and Gemini 3.8 Flash, so this piece is about the two tiers under it). Sol and Luna aren't just cut-down repackagings of last generation's models — OpenAI says both were built using the same methods that improved Astra's professional-work, factuality, coding, computer-use and alignment scores, then tuned for cost and speed instead of raw ceiling performance.
The number every outlet is leading with is the price cut. GPT-6 Sol runs at $2 per million input tokens and $10 per million output tokens, versus GPT-5.6 Sol's $4 and $20 — an exact halving. GPT-6 Luna is priced at $0.10 per million input tokens and $0.50 per million output, down from GPT-5.6 Luna's $0.20 and $1.20 — input cut exactly in half, output cut by more than half (about 58% lower). OpenAI and multiple outlets describe this as permanent list pricing rather than a launch promotion. Sources: Introducing GPT-6 Sol and Luna (OpenAI), gHacks, GPT-5.6 pricing (Finout).
Quick summary: OpenAI launched GPT-6 Sol ($2/$10 per million input/output tokens) and GPT-6 Luna ($0.10/$0.50) on September 22, 2026, roughly halving GPT-5.6-generation pricing. Both share GPT-6 Astra's 1.05M-token context window and reportedly cut internal mistake rates by about half versus their GPT-5.6 predecessors. Luna is now available to ChatGPT's Free and $8/month Go plan users in the desktop app; Sol still requires Plus or higher. Separately, on the same week, GPT-6 Astra was used to help decrypt a WWII German Army Enigma message from July 1941 that had gone unsolved since 2005.
- $2 / $10 per million tokens: GPT-6 Sol's input/output API price — exactly half GPT-5.6 Sol's $4/$20.
- $0.10 / $0.50 per million tokens: GPT-6 Luna's price, down from GPT-5.6 Luna's $0.20/$1.20.
- 1.05 million tokens: the context window shared by Sol, Luna and Astra, with usable input capped around 922K tokens.
- ~56.4%: GPT-6 Sol's reported max-effort score on Agents' Last Exam, a long-horizon professional-tasks benchmark across 55 sub-industries — which OpenAI says tops Claude Opus 5's highest reported score on the same test.
- $10 / $50: GPT-6 Astra's unchanged API price, still the top of OpenAI's now three-tier GPT-6 lineup.
Three tiers, one lineup: where Sol and Luna fit under Astra
OpenAI's GPT-6 family is now shaped like a ladder. Astra ($10/$50 per million tokens) is the reasoning-heavy flagship for the hardest professional and coding work — it's the one LumiChats already put up against Fable 5.1 and Gemini 3.8 Flash. Sol ($2/$10) is pitched as the everyday-work default: more reasoning capability than a pure chat model, aimed at replacing GPT-5.6 Sol as the model most ChatGPT Plus and API users reach for by default. Luna ($0.10/$0.50) is the fast, high-volume tier — built for short, frequent requests rather than deep reasoning chains, the kind of workload where token cost adds up fast. All three now share the same 1.05M-token context window that Astra launched with earlier in September, which is new: previous-generation Sol and Luna tiers didn't always match their flagship's context length.
On accuracy, OpenAI's own internal factuality evaluation — run on de-identified real conversations — reportedly shows GPT-6 Sol making about half as many mistakes as GPT-5.6 Sol, described as approaching Astra-level reliability at a fraction of the cost. Luna's factuality gains are described as substantial too, with OpenAI saying Luna at higher effort settings can match GPT-5.6 Sol's output quality at roughly a hundredth of the cost, though that comparison is OpenAI's own framing and worth treating as a vendor claim rather than an independently verified number until third-party evals catch up. Sources: MarkTechPost, Unite.AI.
| Model | Best for | Input / 1M tokens | Output / 1M tokens | Context window |
|---|---|---|---|---|
| GPT-6 Astra | Flagship reasoning, hardest coding/agent tasks | $10 | $50 | 1.05M (≈922K input) |
| GPT-6 Sol | Everyday work needing more reasoning | $2 | $10 | 1.05M (≈922K input) |
| GPT-6 Luna | Fast, high-volume, low-cost requests | $0.10 | $0.50 | 1.05M (≈922K input) |
Luna's free-tier access: who actually benefits
The part of this launch that matters most for casual users isn't the API pricing — it's that GPT-6 Luna is rolling out to ChatGPT's Free plan and to Go, OpenAI's $8/month low-cost plan (available in India and roughly 170 other countries), inside the ChatGPT desktop app. That's notable because OpenAI usually gates new-generation models behind Plus ($20/month) or higher for weeks after launch. GPT-6 Sol does not get the same treatment: it's arriving for Plus and above, consistent with Sol's positioning as the "default everyday model" for paying users rather than a free-tier upgrade. In practice, this means a Free or Go user who was stuck on GPT-5.5 Instant or an older Luna model gets a meaningfully more accurate, similarly fast model at no extra cost, while Plus subscribers get Sol as a cheaper, more capable stand-in for what used to be GPT-5.6 Sol. Sources: 9to5Mac, ChatGPT Go, OpenAI.
Why the price war, and why now
Sol and Luna didn't land in a vacuum — they capped off roughly 24 hours of competing price cuts across the frontier-model market. Reporting indicates xAI shipped Grok 4.7 at $2/$6 per million tokens on September 21, undercutting prior frontier pricing by a wide margin; Anthropic followed on September 22 with Claude Opus 5.5 at $4/$20; and OpenAI answered with Sol at $2/$10 and Luna at $0.10/$0.50 roughly ninety minutes after Anthropic's move, according to reporting on the sequence. Whatever the exact minute-by-minute order, the practical effect is the same: mid-tier frontier pricing that was $4–$10 per million input tokens a few weeks ago is now largely $2–$4, and that compression is being driven by direct competitive response rather than a single company's roadmap. Sources: OpenAI's GPT-6 Sol and Luna cut prices 50% — Yahoo Finance.
| Model | Maker | Input / 1M tokens | Output / 1M tokens |
|---|---|---|---|
| GPT-6 Sol | OpenAI | $2 | $10 |
| Grok 4.7 | xAI | $2 | $6 |
| Claude Opus 5.5 | Anthropic | $4 | $20 |
| GPT-6 Astra | OpenAI | $10 | $50 |
If you're choosing between tiers for API use: Luna is cheap enough that it's usually worth trying first for classification, extraction, or short-turn chat tasks — Sol is the better default once a task needs multi-step reasoning but doesn't justify Astra's cost.
A separate, unrelated headline from GPT-6 Astra this month: Carter Leffen, a Bloomberg product development coach, used GPT-6 Astra to decrypt "MVUEH," a German Army Enigma-machine radio message sent July 10, 1941, that had resisted every attempt to crack it since at least 2005. The process reportedly took about 10 hours of model time and roughly 14.8 million key checks, with Astra building an Enigma simulator, writing cryptanalysis code, and testing candidate keys in parallel — under human direction, with a researcher supplying the initial crib and steering the investigation rather than the model working fully autonomously. It's not part of the Sol/Luna launch, but it's a useful data point on what Astra's underlying reasoning and coding gains — the same gains Sol and Luna inherit in smaller form — look like applied to an unusual real-world problem.
FAQ
01Is GPT-6 Luna actually free?
Luna is available at no extra cost to ChatGPT Free-plan users and to Go plan users ($8/month) in the desktop app — you don't need Plus or Pro to use it. GPT-6 Sol is not on Free or Go; it requires Plus ($20/month) or a higher tier.
02How much cheaper are GPT-6 Sol and Luna than the last generation?
Sol is priced at $2/$10 per million input/output tokens versus GPT-5.6 Sol's $4/$20 — exactly half. Luna is $0.10/$0.50 versus GPT-5.6 Luna's $0.20/$1.20 — input halved, output cut by roughly 58%.
03Does GPT-6 Sol replace GPT-6 Astra?
No. Astra remains OpenAI's flagship at $10/$50 per million tokens for the hardest reasoning and agentic work. Sol sits below it as the everyday-use default, and Luna sits below Sol for cheap, high-volume tasks — it's a three-tier lineup, not a replacement.
04Are the accuracy improvements independently verified?
Not fully yet. The 'half as many mistakes' figure comes from OpenAI's own internal factuality evaluation on de-identified conversations, not a third-party benchmark, so it's worth treating as a vendor-reported number pending outside verification.
05What was the Enigma story about?
Separately from the Sol/Luna launch, Bloomberg developer Carter Leffen said he used GPT-6 Astra to decrypt a 1941 German Army Enigma message that had gone unsolved since at least 2005, working under human direction over roughly 10 hours. It's an interesting demonstration of Astra's reasoning and coding ability, not a new model release.
If you're trying to figure out whether Sol, Luna, Astra, Grok 4.7, or Opus 5.5 actually fits what you're building or writing, that's exactly the kind of side-by-side LumiChats is built for — you can compare and chat with multiple models in one place instead of juggling five separate apps and pricing pages to find out which one is worth paying for.
