OpenAI Has Just Made Its Most Powerful AI Cheaper. That Matters More Than It Sounds

There was a small update from OpenAI this week that would be quite easy to miss.

On 21 August, OpenAI reduced the API and credit pricing of GPT-5.6 Sol, its flagship model, by more than 20% for the next three months. (OpenAI)

On the surface, it looks like a pricing story.

I think there is something rather more interesting happening underneath it, particularly if you are an author or expert who is beginning to think about turning your intellectual property into something people can actually interact with.

The price of intelligence is continuing to come down.

And that changes what small businesses can build.

There is an important distinction to make first because otherwise this story is very easy to misunderstand. If you have a normal ChatGPT Plus, Pro or Business subscription, OpenAI hasn't suddenly reduced your monthly subscription price by 20%. Those subscription prices and their included usage remain unchanged.

The reduction applies to GPT-5.6 Sol when it is being consumed through the API and to eligible credit-based usage in products including ChatGPT Work and Codex. OpenAI's current rate card lists promotional Sol pricing at $4 per million input tokens and $20 per million output tokens for applicable token-based usage, with the promotion running at least until 21 November. (OpenAI Help Center)

That may sound like a fairly technical distinction, but it matters once we move beyond simply opening ChatGPT and having a conversation with it.

I've been thinking about this quite a lot while writing the AI Book Companion™ Field Guide because there are really two different stages involved when you create something like an AI Book Companion.

The first is proving that the thing should exist.

Does the reader actually use it?

What do they ask it?

Where do they get stuck?

Do they want quick clarification, or do they end up having much deeper conversations with the book?

Does the Companion genuinely help them understand and apply the author's work?

At that stage, I still think there is a very strong argument for keeping things relatively simple.

You don't necessarily need to build an expensive standalone application before you know how people are going to use it. ChatGPT gives you an environment in which you can create the Companion, put it into readers' hands and learn from what actually happens.

The second stage is different.

Once there is evidence that people use it, the question becomes whether you want to build a more integrated experience around it.

Perhaps the Companion sits inside the author's own website. Perhaps access becomes part of a membership or programme. Perhaps the author wants much more control over the reader experience. At that point you are likely to start looking at APIs, hosting and the economics of every conversation that takes place.

And suddenly a 20% reduction starts to matter.

Imagine a Companion that is being used by ten people. The model cost may barely be worth thinking about.

Now imagine 1,000 readers using it.

Or 10,000.

Some will ask one question. Others will spend an hour exploring an idea from the book, coming back repeatedly as they apply it to their own situation.

The commercial question starts becoming quite different.

This is why I am increasingly convinced that authors shouldn't become too attached to whichever AI model happens to be fashionable when they build their first Companion.

The model is infrastructure.

It will get better.

It will get cheaper.

New models will appear.

Some will be stronger at reasoning. Others will be better suited to simple conversations where using the most powerful model available would be unnecessary and expensive.

What needs to remain stable is everything around it.

The author's ideas.

The methodology.

The boundaries.

The language.

The things the Companion is allowed to do.

The things it should never pretend the author has said.

The way it handles questions that fall outside the book.

The reader journey.

That is where the real intellectual property sits.

Once you've built that properly, the underlying AI can change.

I think that is an important distinction because it moves this whole conversation away from building something around ChatGPT, Claude, Gemini or whichever model happens to come next.

You are building an intellectual property system.

The AI happens to be the engine underneath it.

And engines can be changed.

There is another part of this that matters for very small businesses.

For years, sophisticated software had an economic barrier around it. Building something intelligent enough to respond differently to thousands of customers normally meant employing people, developing software or raising fairly substantial amounts of money.

AI is steadily eroding some of that barrier.

A single expert can now take a body of knowledge they have spent ten or twenty years developing and begin turning it into something that can work with people individually.

The economics still need to make sense. There are development costs, maintenance costs and plenty of other considerations that don't disappear simply because the model becomes cheaper.

But the direction of travel is fairly obvious.

The capability is increasing while the unit cost of using it is decreasing.

For authors, I think that will eventually become one of the most important changes AI brings to publishing.

We have spent a lot of time discussing whether AI will write books.

I'm increasingly interested in what happens after the book has been written.

Because if the cost of intelligence keeps falling, a book no longer has to remain a static object that every reader experiences in exactly the same way.

An author's ideas can begin to respond.

And the cost of making that possible has just fallen again.

Three Key Insights

• OpenAI has reduced GPT-5.6 Sol API and eligible credit pricing by more than 20% for a promotional period extending at least to 21 November 2026. (OpenAI)
• The reduction doesn't make ordinary ChatGPT Plus, Pro or Business subscriptions cheaper; its significance is mainly for businesses and developers building services that consume AI through APIs or eligible credits. (OpenAI Developer Community)
• For authors building interactive products around their intellectual property, falling model costs gradually improve the economics of moving from an experimental Companion inside ChatGPT to a more integrated reader experience.

Three Questions & Answers

Does this mean my ChatGPT subscription is now 20% cheaper?

No. OpenAI says Plus, Pro and Business subscription usage is unchanged. The reduction concerns GPT-5.6 Sol API pricing and eligible credit-based usage rather than ordinary subscription fees. (OpenAI Developer Community)

Why would an author care about API pricing?

You probably don't need to care while you're simply using ChatGPT to help with your own work. It begins to matter when you create an AI experience that other people use, because every interaction then has a delivery cost. Lower model pricing makes larger-scale reader experiences more commercially viable.

Does an AI Book Companion™ need to be built as a separate application?

No. There is a strong argument for proving the reader experience first before investing in a larger integration. Once usage and commercial viability are understood, an API-based version can make sense. Falling AI costs make that progression increasingly practical.

Sources

OpenAI
GPT-5.6: Frontier intelligence that scales with your ambition
Pricing update added 21 August 2026.
Full OpenAI source

OpenAI Help Centre
ChatGPT Rate Card — Business, Enterprise/Edu credit-based pricing
Full OpenAI rate card