OpenAI is Rolling Back Pay-as-you-go Pricing for Some Developers
Get ready to pre-pay for access to GPT-4 and other models
OpenAI has begun distributing emails to some developers informing them that they will need to shift from pay-as-you-go end-of-month billing to prepaid token purchases.
Joao Paulo Alqueres, founder of Iara Digital, told Synthedia that he has seven solutions for enterprise customer applications running directly against the OpenAI API and that the billing change was unexpected. On LinkedIn, Alqueres wrote:
Recent changes to OpenAI's API subscription model, transitioning from a pay-as-you-go to a prepaid system, have prompted us to reevaluate our commitment to their services.
This shift presents significant challenges for businesses like ours that rely on flexible billing to scale AI integration in production environments. While we understand the need for companies to evolve their pricing strategies, transparent communication and consideration for existing customers' needs are crucial for sustainable partnerships.
Shifting Users to Channels
This is a big red flag for us having to put 50-dollar credit packages in the platform manually. It’s insane.
OpenAI’s pre-paid billing has been around since at least August. The company’s support page for the service started with a statement that pre-paid billing would become standard for new users but did not mention existing users.
August 23, 2023:
Prepaid billing allows API users to pre-purchase usage. As of today, all new accounts are enrolled in prepaid billing. We're working on opening this billing type up to all users and will also have a feature that allows eligible prepaid billing users to switch to monthly billing…
- Choose the initial amount of credits you want to purchase. The minimum purchase is $5. Currently, the maximum purchase is $50 but we expect to increase this soon!
…
Please note that any purchased credits will expire after 1 year and they are non-refundable…Once you’ve consumed all your credits, your API requests will start returning an error letting you know you’ve hit your billing quota. If you’d like to continue your API usage, you can return to the billing portal and use the “Add to balance” button to purchase additional credits.
The language for this page was updated on December 19th. It removed the information about all new accounts moving to prepaid billing and added a method for automatically repurchasing pre-paid tokens when the credits fell below a specific threshold.
December 19, 2023 - February 8, 2024:
Prepaid billing allows API users to pre-purchase usage.
If you're on a Monthly Billing plan and you choose to purchase credits upfront for API usage, the credits you've bought will be applied to your monthly invoice. This means that any API usage you incur will first be deducted from the prepaid credits. If your usage exceeds the credits you've purchased, you'll then be billed for the additional amount…
- Choose the initial amount of credits you want to purchase. The minimum purchase is $5. Currently, the maximum purchase is $50 but we expect to increase this soon!
Note that the pre-purchase credit limit has not changed since August. This may make you wonder about the definition of “soon.”
The move to pre-payment isn’t entirely different from the pre-paid credits associated with DALL-E or GPT-3 use in the Playground. And you can understand why a company prefers prepayment over payment-in-arrears. This allows them to capture revenue before inference cost is incurred and avoid the inevitable challenges of small-dollar collections. Shifting cash payments forward by 30 or more days may be a tool to help the company capture a one-time cash flow cycle improvement and reduce the perceived need for outside funding.
Shifting users that are consuming hundreds or thousands of dollars per month to prepaid could lead to some complaints, but the impacts are most likely to be benign. It might even incentivize users to migrate to Azure, which is better set up for low-dollar consumption transactions. However, prepaid users hitting an unexpected inference use spike might become a problem.
A credit credit issuer might top-up the OpenAI account a couple or more times in succession. It is almost certain to start rejecting repeated attempts to charge for refilling the prepaid token accounts. This could create runtime user experience problems that go beyond billing inconveniences.
OpenAI’s API is known to have had reliability issues, and the company benefits from revenue flowing through Azure. So, anyone using OpenAI models beyond a test should likely move over to Azure OpenAI Service. The key downside of this is that OpenAI and Azure run different instances of the models. This means users may experience unexpected model behavior in Azure after migration. In addition, OpenAI’s feature updates and pricing are sometimes available through the API before they arrive in Azure.
Refining the Operating Model
OpenAI has scaled to the point of regularly refining its operating model. Given its partnership with Microsoft, it seems logical to incentivize users to shift inference API use to Azure. Granted, this might turn off some customers who do not want to use Azure because they prefer AWS or Google Cloud.
Demand for OpenAI’s models has, to date, led to new Azure signups. However, the developers still using OpenAI’s inference API are likely those who prefer not to use Azure. The billing change may not cause them to abandon OpenAI, but you can see in Alqueres’ LinkedIn post that at least one company plans to evaluate alternative models for its live use cases.
Other OpenAI strategies, such as preemptive price cuts, new model rollouts, and new admin features, are designed to dissuade users from trying other models. The billing change is notable because the attempt to reduce OpenAI’s receivables cycle time runs in the opposite direction.
