How OpenAI’s GPT-6 sol and luna models affect business content and marketing

GPT-6 Sol and Luna are two new OpenAI models that bring much of the capability of its flagship model to everyday work at half the previous price, which makes AI a cheaper option for routine business content and marketing tasks. For companies that want to be found in ChatGPT and Google’s AI Overviews, the same change affects how buyers research suppliers, which is why AI SEO for marketing teams now sits alongside traditional search work. Cheaper models put more content into circulation, so the content that gets noticed has to be accurate and worth citing.

OpenAI set out the details in its announcement introducing GPT-6 Sol and Luna. The company says it trained both models with similar methods to GPT-6 Astra and has cut API prices for Sol and Luna by 50% compared with the promotional pricing for GPT-5.6. It also reports that GPT-6 Sol makes about half as many mistakes as its predecessor on an internal factuality test. The new models are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. The announcement lists no regional restrictions, so UK subscribers on those plans can use them.

What GPT-6 sol and luna are

GPT-6 Sol and Luna are the middle and lower tiers of the GPT-6 family, sitting below GPT-6 Astra. OpenAI describes Astra as its most capable model and says the most demanding projects still call for it, while Sol and Luna cover work that happens at different scales and budgets.

Each model is aimed at a different balance of capability and cost. The three tiers line up as follows.

OpenAI has kept the family moving since launch. A later version, GPT-6.1 Sol, followed at the end of September. The company has since released GPT-6.1 Terra, a fast model for computer tasks and writing that it says is priced 50% below GPT-6.1 Sol through the API.

For a marketing team, the practical point is that the choice of model now matters as much as the choice of tool. Simple tasks can run on a cheaper model while complex research goes to a more capable one.

How much cheaper the new models are

The new models cost much less than the versions they replace. The table below sets out the API prices OpenAI publishes on its model pages, including the lower rate for cached input.

Input tokens are the text sent to the model and output tokens are the text it writes back. Cached input is text the model has already seen in an earlier request, such as a standing set of instructions.

Low cached input prices cut the cost of tools that reuse the same instructions or brand guidelines for every request. For businesses building AI into their own systems, that makes repeated tasks such as product descriptions or email variations cheaper to run at volume.

Most marketing teams use ChatGPT through a subscription rather than the API. For them the change shows up as higher usage limits on the same plan, so more tasks fit within the allowance they already pay for.

Fewer factual errors and a plainer writing style

Sol and Luna make fewer factual mistakes than the models they replace, according to OpenAI’s own testing. The company built its factuality test from conversations where users had flagged errors. It notes that these conversations do not represent typical use.

The models also write differently. OpenAI says it has brought Astra’s communication style to Sol and Luna, with more clarity, less jargon, fewer odd turns of phrase and slightly shorter answers overall.

Both changes help, but neither removes the need for a person to check the output. A model that makes fewer mistakes still makes some. A confident wrong figure in a published article does more damage than a slow first draft. Every statistic, claim and quote still needs a source that a reader can follow.

Plainer writing also means fewer of the tell tale phrases that make AI copy easy to spot. That raises the bar for human editing rather than lowering it, because the remaining problems are harder to see at a glance.

What cheaper AI means for business content

Cheaper and more capable models mean more businesses will publish more content produced with AI help. That increases the volume of similar articles competing for the same searches and the same citations in AI answers.

Volume on its own does not win attention. Content that repeats what every other page already says gives search engines and AI assistants no reason to pick it over the rest. What stands out is first hand experience, original data, clear answers to specific questions and a point of view that only the business publishing it could give.

Our guide to AI content and search rankings looks at how pages written with AI help perform in search. The pattern is consistent: quality and usefulness decide the result, not the tool used to write the first draft.

For a UK business with a small marketing team, the opportunity is to use cheaper models for the groundwork. Research summaries, outlines, first drafts and repurposing existing content all take less time, which leaves more of the budget for the expert input that makes content worth reading.

What Google says about AI content

Google does not penalise content because AI helped to write it, but it does act against content produced at scale to manipulate rankings. Its guidance about AI content in Search says appropriate use of AI or automation is not against its guidelines. The concern is content whose primary purpose is manipulating search rankings.

Google’s spam policies for web search describe scaled content abuse as generating many pages for the primary purpose of manipulating rankings and not helping users. One of the examples given is using generative AI tools to generate many pages without adding value for users.

Cheaper models make that kind of output easier to produce, which is exactly why it carries more risk. A business that publishes hundreds of thin pages because the cost per page has fallen could find those pages working against the rest of the website.

The safer approach is to measure content by what it adds for the reader. If a page would be worth publishing without AI, using AI to produce it faster is fine.

Using AI for marketing tasks beyond writing

GPT-6 Sol and Luna are useful for marketing tasks beyond writing, because OpenAI has improved their ability to use tools and complete work across several apps. The announcement highlights professional work and computer use as two of the main areas of improvement.

For a marketing team, that opens up work such as summarising competitor pages, turning a webinar transcript into a set of social posts, checking a long document for inconsistencies or pulling figures from reports into a summary. These are tasks where a cheaper model can save hours each week without producing anything that goes straight to the public.

The same caution applies to every task. Any output that informs a decision or reaches a customer needs checking by someone who knows the subject. Any figure needs tracing back to its original source.

Choosing the right model for each task also keeps costs down. Luna can handle quick reformatting and simple summaries, while Sol suits research and analysis where accuracy matters more.

Keeping customer data safe when using AI tools

Customer data should only go into an AI tool when the business has a lawful basis and the right agreements in place. The Information Commissioner’s Office publishes guidance on AI and data protection that explains how the principles of UK GDPR apply to information used in AI systems.

Cheaper models make it tempting to paste in a CRM export or a set of customer emails and ask for analysis. Before anyone does that, the business needs to know which plan it is using, what the provider does with the data and whether the use fits its privacy notice.

Personal data is the main risk, but it is not the only one. Pricing, client lists and unpublished plans are commercially sensitive too. The same rules should cover them.

Security is moving quickly as well. OpenAI’s update on the path to releasing Astra says Astra is the first OpenAI model to meet the Critical cybersecurity capability threshold under its Preparedness Framework, which is why stronger safeguards were put in place before release.

How new models change AI search visibility

New models change AI search visibility because they change how ChatGPT reads, summarises and cites the web. A model that makes fewer factual mistakes is more likely to repeat what a source says accurately, which rewards businesses whose websites state the facts clearly.

That makes the accuracy of a business’s own information more important. Our guide to correcting what ChatGPT says about a business explains how out of date or inconsistent information across the web ends up in AI answers.

Structure helps AI tools pick the right answer. The approach behind generative engine optimisation organises content so models can reference and cite it, while answer engine optimisation formats pages as direct answers to the questions buyers ask.

Each model update can shift which sources an assistant prefers. That is a reason to track AI visibility over time rather than checking it once and assuming nothing will change.

A sensible way to test new AI models

The sensible way to test a new AI model is to run it on real tasks the team already does and compare the results with the current approach. Benchmarks published by the model provider are a starting point, but they rarely match the work a marketing team does each week.

A short trial on a handful of tasks shows where the new model saves time and where it still needs close checking. The comparison below sets out what tends to work and what tends to cause problems.

Keeping a record of what was tested and what the team found makes the next model release easier to assess. OpenAI has released several models in a matter of weeks, so the question of whether to switch will keep coming up.

Speed of release is also a reason not to rebuild every process around one model. Workflows that can swap one model for another protect the team from price changes and withdrawn versions.

The lessons from one release carry over to the next. A team that knows how to judge a model on its own work will make faster and better decisions each time.

Getting content right for AI search and Google

Getting content right for AI search and Google comes down to accuracy, expertise and structure, whichever model helps to produce it. Cheaper AI makes the first draft easier for everyone, so the value moves to the parts of content marketing that AI cannot supply on its own.

Every programme Priority Pixels runs combines content audits, content creation and content optimisation through AI workflows its team has designed, built and tested. We measure AI Share of Voice as a companion metric to traditional rankings, which shows where a business appears across Copilot, Gemini, ChatGPT and Perplexity alongside Google and Bing. More detail is on the content marketing service page.

GPT-6 Sol and Luna will not be the last models released this year. Businesses that build content on clear facts, real expertise and regular checks will be in a stronger position whichever model their buyers use next.

OpenAI’s own recent history also shows why oversight matters. Our guide to the Hugging Face incident at OpenAI looks at what happened when AI models went beyond what their testers intended.