Conclusion First: The Debate on Whether to Adopt AI is Over

In July 2025, Alibaba announced that its “Qwen3.8-Max” boasts 2.4 trillion parameters. It accepts text, images, and videos as input and returns text. ChatGPT has become the most utilized AI tool in the U.S. Congress. Meanwhile, AI API prices are collapsing not on a monthly basis, but weekly.

When we put these three pieces of news together, one question emerges:

“Which is currently more expensive: the ‘cost of using AI’ or the ‘cost of not using AI’?

The answer has clearly flipped. This reversal has a devastating impact, especially for local SMEs. Let’s examine the numbers step by step.

—

The Collapse of API Prices — “It’s Too Expensive to Use” No Longer Holds

First, let’s talk about costs.

It’s clear from services like “CostPerPrompt,” which track AI API price trends, that the prices for major models have dropped by 50-80% in the past six months. In 2023, when GPT-4 class models first emerged, the cost was several dozen dollars per million tokens, but by mid-2025, there are multiple models available for under a dollar. Despite being a massive model with 2.4 trillion parameters, Qwen3.8-Max has its API pricing set at a competitive level.

Moreover, this price drop is not a one-time event that stabilizes. It is updated weekly. Prices differ from last week to this week, and they are expected to decrease further next month. This phenomenon should be referred to as “price collapse” rather than just a price drop.

What does this mean in practical terms?

For instance, consider a local e-commerce company that handles 50,000 customer inquiries a month. Traditionally, responding to these inquiries manually requires 3-4 people, including part-timers, costing between 800,000 and 1,200,000 yen per month. If they automate the initial response with an AI chatbot, the API usage fee would be around 10,000 to 30,000 yen per month. What used to cost 100,000 yen six months ago is now less than 30,000 yen. It may drop even further next month.

Understanding this structure can completely change management decisions.

—

The Significance of 2.4 Trillion Parameters — What Can Be Done Has Changed

The sheer number of parameters is irrelevant to SME owners. The real issue is what can now be accomplished with it.

The reason Qwen3.8-Max is gaining attention is its multimodal capability, allowing it to accept not only text but also images and videos as input. What does this mean in practical settings?

• Manufacturing Inspections: By simply sending product photos to the API, initial assessments of defective products can be made without the need to invest hundreds of thousands of yen in a dedicated image recognition system.
• Field Reporting in Real Estate and Construction: By taking photos on-site and sending them to the API, draft reports can be auto-generated. Tasks that used to take site supervisors two hours back at the office can now be completed in five minutes while on the move.
• E-commerce Product Registration: Taking a single product photo can automatically generate product descriptions, specifications, and SEO titles. Tasks that previously took 30 minutes per product can now be done in three minutes.

Areas that previously had “AI capabilities but with questionable accuracy” have now reached practical levels with a model of 2.4 trillion parameters. Moreover, since it can be accessed via API, there’s no need for companies to own GPUs.

For local SMEs, this means that “weapons that only large corporations could afford are now available for a few tens of thousands of yen a month.”

—

What the Fact That ChatGPT is the Most Used AI Tool in the U.S. Congress Indicates

The news that ChatGPT has become the most utilized AI tool in the U.S. Congress may seem like a trivial fact at first glance. However, the essence lies elsewhere.

The important takeaway is that “the most conservative and risk-sensitive organization has integrated AI into its daily operations.”

The applications used by congressional staff — summarizing bills, drafting research reports, responding to inquiries from constituents — are actually quite similar to the daily operations of SMEs. Summarizing documents, drafting emails, and streamlining research.

In other words, excuses like “our industry is still too early” or “this job doesn’t suit AI” are now no longer valid, even in the U.S. Congress.

—

Calculating the ‘Cost of Not Using’ — The Reality for SMEs

Now we get to the main point. The ‘cost of using’ has clearly decreased due to the collapse of API prices. But what about the ‘cost of not using’?

Let’s estimate this for three common cases in local SMEs.

Case 1: Customer Support (E-commerce Company with 5 Employees)

• Average 3 hours per day × 2 people = 6 hours
• At an hourly wage of 1,500 yen × 6 hours × 22 days = approximately 200,000 yen per month
• By automating initial responses with AI, human intervention is only needed for complex cases. With API costs of 10,000 to 20,000 yen per month, 70% of labor costs can be redirected to other tasks.
• Cost of Not Using: 140,000 yen in opportunity loss per month (1,680,000 yen annually)

Case 2: Creating Estimates and Proposals (Construction Company with 10 Employees)

• Sales staff spend 2 hours per estimate. For 20 estimates a month, that’s 40 hours.
• At an hourly wage of 2,000 yen × 40 hours = 80,000 yen in labor costs per month
• By training AI on past estimate data, drafts can be auto-generated, reducing the time per estimate to 30 minutes.
• The saved 30 hours can be redirected to new sales, making it realistic to secure one additional order per month (gross profit of 500,000 yen).
• Cost of Not Using: 500,000 yen in lost sales opportunities per month (6,000,000 yen annually)

Case 3: Recruitment and Job Postings (Manufacturing Company with 30 Employees)

• Outsourcing job posting creation: 50,000 yen per job × 6 times a year = 300,000 yen
• Switching to in-house creation with AI: API costs are less than 10,000 yen annually.
• Cost of Not Using: 290,000 yen in direct cost difference annually
• Additionally, time spent on outsourcing communications (3 hours per job × 6 times = 18 hours) is also reduced.

These are all “conservative estimates.” In practice, when you start, you’ll find many tasks that you might think, “Wait, can AI handle this too?” popping up one after another.

—

A New Habit of Reviewing API Prices Weekly

In a world where AI API prices fluctuate weekly, the “cost estimates made six months ago” are no longer applicable.

SMEs should focus on three actions:

1. Try AI API for One Task Immediately
No need for a grand project. Whether it’s customer support, meeting minutes, or drafting emails — anything will do. You can start for under 10,000 yen a month. There’s a perspective that can only be gained by “getting hands-on” first.

2. Check API Prices at Least Once a Month
Use comparison sites like “CostPerPrompt” to monitor price trends for major models. A model that was too expensive to consider six months ago may now be available at one-tenth the price.

3. Calculate the ‘Cost of Not Using’ Monthly
Inventory the labor costs, outsourcing fees, and time associated with tasks that can be replaced by AI. As long as this figure exceeds the ‘cost of using,’ there’s no reason not to adopt it.

—

The Reversal Structure Unique to SMEs

Finally, let’s discuss the structural aspect.

For large corporations to adopt AI, they must go through security reviews, internal approvals, proofs of concept, and vendor selection — a process that takes at least six months, often a year. During that time, API prices continue to drop, and model performance improves.

SMEs, however, are different. If the CEO decides to proceed, they can start the next day. By obtaining an API key and integrating it into Google Sheets, one task can be automated. This speed of decision-making is the greatest weapon of SMEs.

With a 2.4 trillion parameter model available for a few tens of thousands of yen in API costs, and even the U.S. Congress using AI in daily operations, prices continue to fall weekly.

In this situation, consider calculating the cost of continuing to say “it’s still too early.” That figure could determine your company’s profits for the next year.