GPT-6 Sol and Luna: prices, features and who gets them
GPT-6 Sol and Luna halve OpenAI's API prices. What each model does, per-token costs next to GPT-5.6, who gets them in ChatGPT and what the error claim means.
Source-based. Written from the documents, reporting and reviews linked in the text. Nothing here was tested hands-on by The Ruling Desk. How we work
OpenAI released GPT-6 Sol and Luna on September 22, 2026: Sol is the model for complex work such as coding and agents, and Luna is the cheap, fast one for high-volume simple tasks. Both cost about half as much in the API as the GPT-5.6 models they replace, and both are rolling out in Codex and ChatGPT, with Luna reaching Free and Go users through the desktop app.
Key takeaways
- Sol costs $2 per million input tokens and $10 per million output tokens; Luna costs $0.10 and $0.50. GPT-5.6 Sol was $4 and $20, and GPT-5.6 Luna was $0.20 and $1.20.
- Cached input is billed at 10% of the normal input rate, so repeated context costs $0.20 per million tokens on Sol and $0.01 on Luna.
- In ChatGPT, paid plans get both models in ChatGPT Work and Codex. Free and Go users get Luna only, in the desktop app. Neither model is in the regular chat view yet.
- "About half as many mistakes" is OpenAI's own claim, measured on an internal factual-accuracy test built from ChatGPT conversations where users flagged errors. It is not a coding benchmark, and nobody outside OpenAI has checked it yet.
What GPT-6 Sol and Luna are for
With this launch announcement, OpenAI now splits GPT-6 into three tiers. GPT-6 Astra, which came out earlier in September, is the most capable and most expensive. Sol sits in the middle: OpenAI's model page for GPT-6 Sol says it is "built for complex coding and agentic workflows," meaning tasks where the model plans, calls tools and works through several steps on its own.
Luna is the budget tier. OpenAI calls it its "most efficient model for focused, high-volume tasks", and TechCrunch's launch story quotes the company's examples: summarizing documents, extracting information and answering quick questions.
On paper, the two share the same limits, according to OpenAI's documentation:
- A context window of 1,050,000 tokens (how much text the model can read at once) and up to 128,000 output tokens.
- Six reasoning levels, from
nonetomax, withmediumas the default. - Text and image input, text output, plus web search, file search, code interpreter and computer use through the Responses API.
- Knowledge cutoffs of April 20, 2026 for Sol and May 18, 2026 for Luna.
GPT-6 Sol and Luna prices next to GPT-5.6
These are standard API prices in US dollars per million tokens, as listed on OpenAI's pricing page on September 23, 2026. GPT-5.6 Luna no longer appears there, so its old price comes from VentureBeat's launch coverage.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | $10.00 | $1.00 | $50.00 |
| GPT-6 Sol | $2.00 | $0.20 | $10.00 |
| GPT-5.6 Sol | $4.00 | $0.40 | $20.00 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
| GPT-5.6 Luna | $0.20 | n/a | $1.20 |
So the GPT-6 Sol price is exactly half of GPT-5.6 Sol's on both input and output. Luna's input is also half, and its output drops by about 58%. VentureBeat reports that OpenAI calls these permanent prices, not a launch promotion.
The fine print matters if you build on the API. OpenAI's model pages say prompts over 272,000 input tokens cost 2x on input and 1.5x on output for the whole request, and writing to the cache costs 1.25x the input rate ($2.50 per million on Sol). Batch and Flex processing run at half the standard rates; Fast mode costs double.
Who gets GPT-6 in ChatGPT
The ChatGPT rollout is narrower than "live in ChatGPT" suggests. According to Techzine, Sol and Luna are available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and the models "aren't available in Chat yet," meaning the regular conversation view most people use.
For free users, the news is Luna. People on ChatGPT Free and Go can use it in the ChatGPT desktop app, while Sol stays with the paid plans. TechCrunch says the rollout to the ChatGPT app and website was gradual through launch day, so some accounts may not see the models right away.
For developers, both are live now in the API as gpt-6-sol and gpt-6-luna, through the Responses, Chat Completions and Batch endpoints.
What "half as many mistakes" actually measures
OpenAI says Sol makes about half as many mistakes as GPT-5.6 Sol and approaches the reliability of its top model, Astra. Per TechCrunch, the figure comes from a "factuality evaluation" built from de-identified real-world ChatGPT conversations in which users had flagged mistakes.
That is a useful test, but it is a narrow one. It measures factual errors in chat answers, not bugs in code, and OpenAI built it, ran it and reported it. No independent lab has published a replication as of September 23, 2026, so treat "half" as the company's number, not a settled fact.
OpenAI also published safety figures, as reported by VentureBeat. On a coding-deception test, the company says Sol misled users 1.3% of the time, down from 10.4% for GPT-5.6 Sol. These are OpenAI's internal results too.
How Sol and Luna compare with rival models
The benchmark claims are OpenAI's, reported by VentureBeat. On AutomationBench, OpenAI says Sol at its highest effort scored 33.2%, ahead of Anthropic's Claude Opus 5 at 26.9%, for about 27 cents per task. On the DeepSWE coding benchmark, OpenAI puts Sol at 68.8% and Luna at 66.6%, just under Claude Fable 5 at 69.9%.
Those comparisons are already a step behind. Anthropic released Claude Opus 5.5 the same day with its own price cut, and OpenAI's charts don't include it. Our Claude Opus 5.5 breakdown covers what Anthropic charges and claims, and the AI section collects the rest of our model coverage.
What happens next
Open questions to watch:
- Regular Chat access. Neither model is in ChatGPT's main chat view yet, and OpenAI hasn't given a date.
- Independent tests. Outside benchmarks will show whether the error and coding claims hold up beyond OpenAI's own evaluations.
- GPT-5.6 Sol's future. It is still listed on OpenAI's pricing page, which says its promotional pricing runs at least through November 21, 2026.
Bottom line
If you pay for the API, GPT-6 Sol and Luna are a straightforward cut: half the price of GPT-5.6 Sol, and half or less for Luna. If you use ChatGPT for free, the change is Luna in the desktop app, not Sol. The "half as many mistakes" line is OpenAI's own claim, measured on factual answers in its internal test, so wait for independent results before counting on it.
FAQ
Is GPT-6 Luna free to use?
In ChatGPT, yes, for Free and Go users, but only in the desktop app for now. In the API, Luna costs $0.10 per million input tokens and $0.50 per million output tokens.
How much does GPT-6 Sol cost?
In the API, $2 per million input tokens, $0.20 per million cached input tokens and $10 per million output tokens, as of September 23, 2026. Prompts over 272,000 input tokens cost more. In ChatGPT, Sol comes with the Plus, Pro, Business, Enterprise and Edu plans.
What is the difference between GPT-6 Sol and Luna?
Sol is the stronger model, meant for coding and multi-step agent work. Luna is 20 times cheaper per token and built for simple, high-volume jobs like summaries and data extraction. Both have the same context window and tools, according to OpenAI.