Conclusion

Let’s get straight to the point. Small and medium enterprises that have not yet decided how much they will pay for AI each month should make that decision by the end of this month.

In the past week, the pricing structure of AI has undergone significant changes.

Apple has introduced AI features in Logic Pro as “Intelligence Usage Packs,” charging users based on their usage. Anthropic announced its new model, Claude 4 (Sonnet/Opus), which has nearly halved inference costs compared to the previous generation. OpenAI has also continued to lower prices for GPT-4o.

What this trend signifies is simple. AI has shifted from a binary choice of ‘free or paid’ to a gradient of ‘how much to spend each month.’

And this is nothing but a tailwind for small and medium enterprises. Why? Because it has become an era where they can acquire “usable tools” within a monthly budget of tens of thousands of yen, without needing to allocate AI budgets in the tens of millions like large corporations.

However, if they continue to incur charges without setting a cap, they might find themselves exceeding 100,000 yen a month before they know it — this has already started to happen. Therefore, what needs to be done this month is one thing: decide how much we will spend on AI each month.

—

What Happened: Interpreting Three News Items as Changes in Cost Structure

1. Apple: Turned AI into a Pay-Per-Use ‘Function Pack’

Apple has started selling the AI features of Logic Pro (music production software) as separate “Intelligence Usage Packs” at a price of $49.99 (about 7,500 yen) per pack. Users can purchase only what they use, such as track generation and mixing assistance.

The essence of this move is that “Apple has passed the computational costs of AI onto the users.” Even a company as large as Apple can no longer include the costs of AI inference in their subscription monthly fees. Conversely, the cost of AI features is at a delicate balance where it is ‘not cheap enough to give away for free’ but ‘has become cheap enough to sell individually.’

Implication for small and medium enterprises: We have entered an era where AI features can be purchased a la carte rather than through an expensive all-inclusive plan.

2. Anthropic: New Model Halves Inference Costs

Anthropic’s newly announced generation model (Claude Sonnet 4 / Opus 4) has significantly improved coding performance while reducing API inference costs by about 40-50% compared to the previous generation.

Specifically, the cost for generating 1 million tokens with Claude 3.5 Sonnet was $15, but with Claude Sonnet 4, users can achieve similar or better results at a lower cost. The monthly Pro plan remains at $20 (about 3,000 yen), while the Max plan is $100 (about 15,000 yen).

What does this mean? AI utilization that was deemed ‘too expensive to consider’ six months ago can now be tested at half the price. The “break-even point” for small and medium enterprises’ AI budgets is continually decreasing.

3. Accelerating Price Competition: The Triad of OpenAI, Google, and Anthropic

Let’s summarize the monthly fees for major AI services aimed at individuals and small businesses.

Notably, the Pro plans of the three major companies are all aligned at ‘monthly $20 = about 3,000 yen.’ Six months ago, just being able to use GPT-4 for $20 a month was groundbreaking. Now, at the same price, users can choose from the latest models of all three companies.

The ‘unit price’ of AI is definitely decreasing. The question is how to utilize the reduced costs.

—

How to Determine the ‘Monthly AI Spending Cap’ for Small and Medium Enterprises

First, Stop Thinking in Terms of Revenue Ratios

The common notion of “investing a certain percentage of revenue in IT” does not apply to AI. This is because the monthly costs of AI tools, ranging from 3,000 yen to 20,000 yen, are determined by the decision of whether to use them or not, regardless of the scale of revenue.

It is nonsensical for a company with annual sales of 30 million yen to judge a 3,000 yen monthly ChatGPT Plus subscription based on a revenue ratio. That’s a discussion about coffee expenses, not investment decisions.

The Correct Approach: Can You Recoup Costs Based on ‘Time Saved by AI × Hourly Rate’?

This is the only criterion small and medium enterprises need to use when deciding on their AI budget.

“How many hours of work can be saved with this AI tool each month? Can the savings be recouped at the hourly rate?”

Let’s look at some specific examples.

• ChatGPT Plus (3,000 yen/month): Used for drafting emails, summarizing meeting minutes, and research, saving 5 hours a month → Equivalent to 10,000 yen at an hourly rate of 2,000 yen. Profitable.
• Claude Pro (3,000 yen/month): Used for contract checks and creating long manuals, saving 3 hours a month → Equivalent to 6,000 yen at an hourly rate of 2,000 yen. Profitable.
• Gemini Advanced (3,000 yen/month): Used for data organization with Google Sheets, saving 2 hours a month → Equivalent to 4,000 yen at an hourly rate of 2,000 yen. Profitable.

In other words, an AI tool costing 3,000 yen a month can pay for itself if it saves at least 1.5 hours of work per month (at an hourly rate of 2,000 yen). If this cannot be achieved, the free tier is sufficient.

Practical Guidelines for Setting a ‘Spending Cap’

Instead of revenue scale, determine based on the number of people using AI and the depth of usage.

The last category, “integrating into business systems,” actually presents the biggest opportunity for small and medium enterprises. Previously, developing business systems would cost 3 to 5 million yen, but now, using AI APIs, there are increasing cases where it can be managed with a running cost of about 50,000 yen a month. Automatic creation of estimates, automatic classification of inquiries, anomaly detection in inventory data — these “mundane but daily tasks” are where AI can be most effective.

—

“So, what should we do?” — Three Things to Do This Month

1. First, sign up for a paid plan.

Whether it’s ChatGPT Plus, Claude Pro, or Gemini Advanced, it doesn’t matter. It’s 3,000 yen a month. Too many people are judging AI based on the free tier, thinking, “Is this all there is to AI?” The paid versions are a different beast. Use it to its fullest for a month.

2. Record ‘how many hours you saved each month.’

Measure it in numbers, not feelings. If you save 15 minutes on email creation × 20 business days = 5 hours a month, this figure will provide a basis for continuing next month. If not, you can cancel. After all, it’s just 3,000 yen a month.

3. Set a spending cap ‘by the end of this month.’

AI tools will continue to proliferate. If you sign contracts for “this and that” every time a new model is released, you may find yourself exceeding 50,000 yen a month before you know it. Set a cap first, then choose the most effective tools within that limit. This is the correct approach to AI investment for small and medium enterprises.

—

The Real Issue: What Happens After AI Costs Decrease?

Finally, let’s delve a bit deeper.

Apple has turned AI into a pay-per-use model, and Anthropic has halved inference costs. If this trend continues, it is highly likely that prices will be further halved in a year.

What will happen then? “Using AI” will no longer be a differentiator.

In an era where everyone uses AI, the competition will hinge on “how to reinvest the time and costs saved by AI.” Large corporations will reinvest the saved costs back into AI development. But what about small and medium enterprises?

The answer is to “focus on things that only humans can do.” Building relationships with local customers, identifying challenges that can only be understood on-site, and adding a touch of “our uniqueness” to drafts generated by AI — these are tasks that a 3,000 yen AI cannot accomplish.

AI is a tool. If the price of tools has decreased, buy them quickly and use the saved time for “human work.”

That is the only strategy small and medium enterprises should adopt in the AI era.