How should OpenAI price o1?

Steven Forth is CEO of Ibbaka. Connect on LinkedIn

OpenAI has released o1, a new model with reasoning abilities. It is meant to be a quantum leap forward in the capabilities of the foundation models that underpin generative AI applications. You may have first heard of it as Strawberry, or even as Q*. It is available to some ChatGPT and Microsoft Copilot accounts and you can expect it to become widely available over the next few weeks. Plan to start using it as soon as you can.

You can take the survey here, we will keep it open until Sept. 19 | OpenAI o1 is already starting to show up in the survey results!

What and why o1 and why should people in pricing care?

Large Language Models are known to be relatively poor at math and complex logical reasoning. This is a result of their being trained on a vast amount of content where there are many errors and reasoning shortfalls. There have been two treatments for this. One is to recognize when real math is needed to answer a prompt and to direct the system to an intelligence that is capable of answering mathematics questions like Wolfram Alpha. The second is to structure a sequence of prompts better using the ReAct (Reason Action) approach and chain of thought style prompting hoping that this will lead to better results or at least clarify where errors are likely to be found.

OpenAI o1 seems to be taking both approaches and in terms of chain of thought has raised things to a whole new level. The result, in Open.ai’s words, is

“We trained these models to spend more time thinking through problems before they respond, much like a person would. Through training, they learn to refine their thinking process, try different strategies, and recognize their mistakes. 

In our tests, the next model update performs similarly to PhD students on challenging benchmark tasks in physics, chemistry, and biology. We also found that it excels in math and coding. In a qualifying exam for the International Mathematics Olympiad (IMO), GPT-4o correctly solved only 13% of problems, while the reasoning model scored 83%. Their coding abilities were evaluated in contests and reached the 89th percentile in Codeforces competitions..”

See Introducing OpenAI o1 Preview and Learning to Reason with LLMs.

Ibbaka is beginning to use o1 for things like scaling tests on value - price - cost models (remembering that to be viable these must satisfy relationship Value > Price > Cost across scale) and an evaluation and recommendation system for value models and value stories.

Community thoughts on pricing o1

We did a simple poll on LinkedIn to ask our community how Open.ai should price o1. The poll data is from Thursday, Sept. 12 to Saturday, Sept. 14. There were 82 responses distributed across the Artificial Intelligence Exchange: AI Conversations and Collaborations group, the Professional Pricing Society group, the Design Thinking group, the Software as a Service - SaaS - Group and Steven Forth’s personal feed.

The question was:

“ChatGPT and o1 (Strawberry) should be priced …”

• At a large premium

• At a small premium

• No premium

• As a separate offer

From LinkedIn poll "ChatGPT + Strawberry should be priced ..." data from Sunday Sept. 15, 2024 N = 82

The results were pretty similar across the different groups (which is somewhat unusual for these polls), with a small preference in the SaaS group for a separate offer.

Strategy and pricing tradeoffs for o1

How OpenAI prices o1 will depend on the role it is meant to play in their portfolio. There are several possibilities here (not mutually exclusive).

• Reinforce its dominance in foundation models and ensure that as many solutions as possible are built on Open.ai models

• Sharpen the differentiation between Open.ai models and other foundation models (this supports the dominance strategy but is not the same as one can assert dominance by having an undifferentiated offer and one can differentiate for niche markets without seeking dominance

• Create new data for training future models - OpenAI has said that this is part of the plan

• Support a modular architecture that is more flexible and faster to adapt to a rapidly changing environment

• Demonstrate value for specific applications as there has been pushback in some quarters on the value being delivered by ChatGPT and Microsoft Copilot.

• Provide the value needed to justify a price increase as OpenAI continues to hemorrhage money and is not driving enough revenue per user

There are conflicting reports on the value that users are getting from generative AI. The naysayers have been getting the most media attention, but we have seen some yet-to-be-released reports that contradict this, especially when CIOs and CTOs are surveyed.

OpenAI is driving significant revenues. The most recent estimate by The Information is that OpenAI has about 11 million paying users and is driving more than US$225 million per month (or $2.7 billion per year).

Pricing depends on strategy, so how OpenAI prices o1 will depend on where it sees it fitting into a larger strategy. We summarize this below.

It appears that the current decision is to package o1 into as many applications as possible at no additional cost. As a horizontal application is a race for market dominance with Anthropic, Google, Meta, Mistral, and the buzzing crowd on Hugging Face this makes sense. OpenAI will be able to capture the added value of o1 as it begins to get used in vertical AI applications (see Bessemer The future of AI is vertical).

Let’s map the strategic options to the pricing choices (remembering that one can have more than one strategic option and that with smart packaging one can execute on more than one pricing strategy.

OpenAI seems to have made the Reinforce Dominance and Create Data strategic choices and as a result, we should not expect them to change pricing just because they are introducing o1.

There is another cluster of strategic choices that would lead to the creation of a separate offer: Increase Prices, Sharpen Differentiation, Modular Architecture, and Demonstrate Value. It is possible that OpenAI will execute this through partners, not so much Microsoft, which is caught in the same games as OpenAI and Google, but the many companies developing vertical AI applications.

It is possible that OpenAI is waiting until a more mature version of o1 before it tests other pricing models and pricing levels. See OpenAI releases Strawberry reasoning model to paying customers.

Note that there is no strategic reason for a small price increase.

Please contribute and share the survey with your community.

Book a demo today and discover how our unique B2B SaaS solution can increase sales, improve retention, and drive growth.