Last month’s estimates are already outdated this month.

ChatGPT has surpassed 1 billion weekly active users. The successor to GPT-4o, GPT-5.6, has further reduced API costs. An OSS tool has emerged that cuts PDF token costs by up to 99%. All three of these developments are happening simultaneously.

For small and medium-sized business owners, this is not just a story of ‘AI is amazing.’ The cost structure itself is being rewritten on a weekly basis.

—

What 1 Billion Users Means—The ‘Non-Users’ are Becoming the Minority

In May 2025, OpenAI announced that ChatGPT’s weekly active users had surpassed 1 billion. This is a significant increase from about 400 million at the end of 2024, representing a 2.5-fold growth in just six months. It is now on par with Gmail and YouTube in scale.

What this number indicates is the fact that ‘AI is no longer something special.’ Clients, competitors, and job seekers are all using AI as a matter of course. In other words, the timing is already here when companies that do not use AI will be seen as ‘behind the curve.’

Even in local small and medium enterprises, the number of companies starting to use ChatGPT for tasks such as creating estimates, summarizing meeting minutes, and drafting responses to inquiries is definitely increasing. The issue has shifted from ‘whether to use it’ to ‘how to use it to minimize costs and maximize results.’

—

The Shock of GPT-5.6—Performance Increases While Costs Decrease

OpenAI announced GPT-5.6 (codenamed: Quasar) in May 2025. What stands out is not the specifications but the change in price structure.

According to the official announcement of API usage fees, the cost per input token has decreased by about 33-50% compared to GPT-4o. This means that even for the same processing, the billing amount could be reduced to two-thirds or half.

Let’s consider this concretely:

• Suppose a small to medium enterprise uses 5 million tokens via API each month.
• The processing that cost about 50,000 yen per month based on GPT-4o would drop to about 25,000 to 33,000 yen with GPT-5.6.
• This translates to a difference of 200,000 to 300,000 yen annually.

You might think that a few hundred thousand yen is not a big deal. However, for small and medium enterprises, an annual saving of 300,000 yen represents a budget for ‘automating another business with AI.’ With the saved costs, they can automate invoice processing or semi-automate recruitment screening—such reinvestments are possible.

Moreover, this is the situation as of this month. OpenAI has updated its models every six months to a year over the past two years, consistently lowering costs each time. It has become the norm to think, ‘It might get cheaper next month.’

—

99% Reduction in PDF Tokens—The Era of ‘Just Reading Costs Thousands of Yen’ is Over

Another significant impact on the ground is the drastic reduction in the cost of PDF processing.

The OSS tool ‘TokenSaver’ (and similar local RAG tools) published on GitHub can reduce token consumption by 90-99% when processing large documents like PDFs with AI.

The mechanism is simple. Instead of sending the entire text of the PDF directly to the API, it extracts text locally, chunks it, and performs vector searches, sending only the relevant parts of the questions to the API. Even for a 200-page contract, only a few relevant paragraphs are actually passed to the API.

Here’s how the numbers break down:

Annual costs drop from 1 million yen to below 100,000 yen.

This means that the costs for AI to process the ‘mountain of paper’ that local small and medium enterprises handle daily—such as checking construction specifications, reviewing real estate contracts, and searching technical documents in manufacturing—have fallen to practical levels.

The excuse of ‘I want to let AI read it, but the costs don’t add up’ is no longer valid.

—

The Break-Even Point for the ‘50,000 Yen Monthly Era’ is No Longer Fixed

When we layer these three changes, a structure becomes visible.

The break-even point for the cost of AI implementation is continuously decreasing on a weekly basis.

A business that was deemed too costly at ‘50,000 yen per month’ a month ago might now be feasible at 20,000 yen this month. It could drop even further next month.

In this situation, small and medium enterprises should take three actions.

1. Avoid Easy Lock-In to Annual Contracts or Fixed Plans

Even if a SaaS AI tool offers ‘20% off for annual contracts,’ the likelihood of cheaper options emerging within six months due to model generational changes is high. Maintain a monthly contract or pay-per-use model to ensure flexibility in switching.

2. Have the Option of ‘Direct API Access + OSS’

While some tasks may be adequately handled with ChatGPT Plus at $20 per month (about 3,000 yen), there are also tasks that are significantly cheaper when integrated into the company’s workflow via API. By combining OSS tools like TokenSaver, there are increasing cases where the same work can be done for less than paying several thousand yen per month for SaaS.

Even without in-house engineers, simply consulting local IT supporters or freelancers about ‘can we set this task up with API + OSS?’ can sometimes halve the costs.

3. Make ‘Small Trials and Weekly Reviews’ a Habit

Reflecting on AI costs in monthly management meetings is too late. Weekly checks on ‘the number of tokens used this week,’ ‘the volume of tasks processed,’ and ‘the labor hours saved’ should be conducted. The numbers can be found on the API dashboard.

The key is to shift the mindset to manage AI costs as ‘variable costs’ rather than ‘capital investments.’

—

So, What Should We Do?

For local small and medium enterprises, the current situation can be summed up in one sentence.

The ‘cost of not using AI’ has begun to exceed the ‘cost of using AI.’

With 1 billion people using ChatGPT, competitors and clients are starting to operate with AI as a premise. With model price cuts and the evolution of OSS tools, there are increasing tasks that can be initiated for just a few thousand yen a month.

There is no need to build an AI infrastructure costing several million yen like large corporations. In fact, small and medium enterprises are at an advantage due to their faster decision-making. If a new model comes out next week, they can switch to it immediately. While large corporations are going through approval processes, small businesses can start using it right away.

This ‘agility’ is the greatest weapon that small and medium enterprises have in the AI era.

Start by selecting one task in your company that is currently consuming the most time with ‘paper and Excel,’ and try throwing it to ChatGPT today. The cost will be just a few dozen yen. If it doesn’t work out, you can stop it tomorrow.

The break-even point will shift again next week. There’s no longer a reason to wait.