Clay illustration: a huge ornate engine carrying one parcel beside a small efficient engine moving many parcels, showing a cheaper AI model doing more work

Quick answer

For most everyday business work, you probably do not need the most expensive AI model. In a new video, Nate B Jones argues that many jobs, from customer support replies to routine coding and drafting, are now “saturated”: several models already do them well enough, so paying for extra intelligence adds little. Companies are testing cheaper models against their own real tasks and keeping the cheapest one that passes. The result is more AI use without a bigger bill. The practical takeaway: test before you upgrade, count the full cost (including checking and fixing the output), and save premium models for the hard cases that genuinely need them.

Key takeaways

  • “Fantastic” and “necessary” are different things. A frontier model can be brilliant and still not improve your particular task enough to justify the price.
  • Tests beat hunches. Nate’s enterprise contacts pick a cheaper model and check it against simple tests of their real work, called evals.
  • Usage can grow 5x on the same budget. Moving routine work from $5 to $1 per million tokens buys five times the work for the same spend.
  • Spending data backs the trend. Ramp’s November AI Index shows business AI spend up just 1.65% in October while token prices keep falling.
  • Your customers will do the same maths. If AI makes your service cheaper to deliver, expect clients to ask for some of the savings.

🛠️ Practical guides

Ready to put this into practice? These step-by-step guides answer two common follow-up questions:

Every few weeks a new “best ever” AI model arrives, and it is natural to wonder whether you should be paying for it. The honest answer for most small businesses is: only if your own work proves you need it.

That is the heart of Nate B Jones’s latest video, and it is a useful reality check. In this article we sum up his argument about the most expensive AI model and why many companies are trading down, check his main figures against public sources and turn it into steps you can use this week. If you want the model-by-model detail, our guides to choosing a Claude model and comparing AI models on token cost go deeper.

Watch: the source video

What this video covers: Nate B Jones explains task saturation, why enterprise buyers now use evals to pick the cheapest model that does the job, what Ramp’s flat AI spending data suggests, when frontier models still earn their price and why your own customers will soon ask you to share your AI savings.

Video: Nate B Jones (@NateBJones), “Why Your Company Doesn’t Need the Most Expensive AI Model”, published 15 November 2026. No paid sponsor is mentioned; Nate promotes his own Evals Kit and Substack newsletter. This summary, the fact-checks, the opinions and the practical advice are our own.

What is task saturation?

Short answer: task saturation is the point where several AI models can already do a particular job well enough, so buying a smarter, pricier model adds little or nothing to the result.

Nate’s point is not that AI has stopped improving. The top models keep getting better. It is that a lot of ordinary business work does not need that improvement. A polite reply to a delivery question, a first draft of a product description or a tidy meeting summary has a ceiling. Once a model reaches it, extra intelligence is wasted money.

He gives a telling example: people in his community have spent hundreds of dollars running a strategy document through top models, then wondered whether the result was any better than a cheaper option would have produced. The model was impressive. The decision did not improve enough to justify the bill.

💡 Tip: A quick way to spot a saturated task: if you could hand it to a capable junior team member with a good checklist, a mid-priced model can probably handle it too.

Why are businesses choosing cheaper AI models?

According to Nate, the conversations he has with large companies have changed in the last few months. They no longer ask him which model to use for which job. Instead they:

  1. pick a relatively cheap model,
  2. test it against evals, which are simple tests of the real work they need done, judged against an agreed standard of “good enough”,
  3. and keep moving towards the cheapest model that still passes.

This matters because it turns “I think we are overpaying” into evidence. If a cheaper model passes the same tests, you can prove it is a fair substitute. The models do not need to be equally clever at everything. They only need to meet your requirements.

That, Nate argues, shifts pricing power from the AI companies towards buyers. When several models pass your test, suppliers have to compete for work you can move quickly. You do not even need to switch supplier: most AI companies sell a range of models, so you can simply move down their price list.

How can AI use grow while the bill stays flat?

Nate uses a simple worked example. We have laid it out below (his figures, in US dollars, chosen for easy maths rather than real prices).

BeforeAfter moving routine work to a cheaper model
Average price per million tokens$5$1
Tokens used in the month1 billion5 billion
Monthly bill$5,000$5,000

The business uses five times as much AI and pays the same. The extra 4 billion tokens are doing useful work, but the model maker earns nothing extra. For you as a buyer, that is good news: you can find more jobs for AI without your costs climbing at the same rate.

Tokens, by the way, are the small chunks of text a model reads and writes. Most AI tools that charge by usage price them per million.

⚠️ Warning: A cheaper price per token is not always a cheaper job. If the lower-cost model needs three attempts, or your team spends an extra half hour fixing its work, the saving can vanish. Always count review and correction time, which is exactly what our guide to testing a cheaper model walks you through.

What does the spending data actually show?

Nate leans on the Ramp AI Index, which tracks AI spending by more than 85,000 US businesses that use Ramp’s company cards and bill payments. We checked the November 2026 report, published on 11 November:

  • Business AI spending rose just 1.65% in October, a new high but very slow growth.
  • 56.8% of businesses in Ramp’s sample now pay for AI, up one percentage point.
  • The average price per million tokens fell 21% in a month, and token volumes keep growing faster than token spending.
  • Customer success and support teams went from 0.8% of classified AI spend in June to 13.3% in October.
  • Frontier models still took 69.5% of spending, up slightly from 67.7%.

Nate also says Ramp’s October report showed almost no growth in token spend across August and September. Put together, he reads this as three months of broadly flat spending while usage keeps rising.

🧾 Evidence vs opinion: The Ramp figures above are published data. The idea that flat spending signals a lasting shift of value away from AI model makers is Nate’s interpretation. Ramp itself notes its customers skew towards tech-savvy, growth-focused US firms, and free tools or personal accounts do not show up at all, so treat it as a strong signal rather than the whole market.

When is the most expensive AI model worth it?

Short answer: when the task is genuinely hard and getting it wrong is costly.

Nate is clear that there is still a premium market. The fact that frontier models take nearly 70% of spending shows some work justifies the price. Examples where the top model may earn its keep:

  • complex, multi-step reasoning such as a tricky contract review or financial model,
  • difficult code changes in a large, messy codebase,
  • the small share of customer queries that are unusual, emotional or high-stakes.

His suggested pattern is routing: a cheaper model handles most of the volume and only the hard cases go to the expensive one. Many AI tools now include a router or an “auto” model setting that does this for you. Even if a cheaper model needs two or three tries, Nate says it can still work out cheaper than one pass with the priciest option. Your tests will tell you which.

Type of taskLikely model tierWhy
Routine replies, summaries, tidy-upsCheaper or mid-tierSaturated: many models already do it well
First drafts of posts, emails, descriptionsMid-tierGood enough quality, you edit anyway
Hard reasoning, high-stakes documentsTop tierMistakes are expensive; extra quality shows
Mixed workloads (support inbox, coding)Router: cheap by default, top tier for hard casesPays premium only where it helps

If your tool lets you choose how hard a model “thinks”, that is another lever. Our guide to choosing an AI reasoning effort level covers it.

Will your own customers ask for AI savings?

This is the part of the video most relevant to service businesses. Nate points out that your customers will apply the same logic to you.

He cites Infosys, the large IT services firm. Its October results showed AI revenue growing fast (29% quarter on quarter, now 10.5% of revenue) while clients pushed for AI productivity gains to be passed through as lower prices. Its operating margin held at about 21%, but only through hard work on efficiency.

Nate’s own example: you charge $10,000 for a service that costs $6,000 to deliver, leaving $4,000 margin. AI brings delivery costs down to $4,000. For a while you might keep the extra $2,000, but if a competitor can deliver the same result for less, you will feel pressure to cut your price. He calls that a healthy pressure: it pushes you to show real value on top of the AI.

If you sell services, our guide on what to say when a client asks for a lower price because you use AI helps you prepare for that conversation.

What should you invest in that lasts?

Nate’s closing advice is about what survives the next model upgrade. Some investments age badly: complicated workarounds built to cover a model’s weaknesses often become unnecessary when a better model arrives. He has seen several this year that had to be simplified.

Other investments get more valuable as models improve:

  • Organised customer information that any model can use,
  • Connected systems, so AI can reach your inbox, CRM or orders safely,
  • A reliable way to check results, such as a small set of tests for your key tasks.

This links neatly to ownership. Your data, your processes and your quality checks belong to you and move with you between tools. A clever prompt hack tied to one model does not.

What should a small business do now?

Here is a calm, practical plan that does not need a procurement team.

  1. List your regular AI jobs. Write down the five tasks you use AI for most, and which model or plan each uses.
  2. Mark the saturated ones. Which are routine, repeatable and easy to check? Those are your test candidates.
  3. Run a small test. Give a cheaper model five to ten real examples and judge them against the same standard. Our step-by-step testing guide shows how.
  4. Count the full cost. Include your time spent checking and fixing, not just the subscription or token price.
  5. Keep the premium model for hard cases. Use a router or simply switch manually when the task is tricky.
  6. Review your plans. If cheaper models cover most of your work, you may not need the top subscription for everyone. Check renewal dates and see our guide to annual or monthly AI plans.

Frequently asked questions

Do I need the most expensive AI model for my business?

Usually not for everyday work. Test a cheaper model on your real tasks first. Keep the top model for hard, high-stakes jobs where the extra quality clearly shows.

What is an AI eval?

An eval is a simple test of the work you actually need done: a set of real examples plus an agreed standard for a good answer. You run each model through it and compare results and costs.

What does task saturation mean?

It means several AI models can already do a task well enough that paying for a smarter model adds little or nothing to the result.

Is AI spending really flat?

Ramp’s November 2026 AI Index shows business AI spend rose only 1.65% in October while token prices fell 21% in a month. That suggests more use for similar money, though Ramp’s sample is mostly US businesses.

Will cheaper models be worse at my tasks?

Sometimes. That is why you test. If a cheaper model fails your test or needs a lot of fixing, the expensive one may be worth it for that task.

Should I cut my prices because I use AI?

Not automatically. Expect clients to ask, and be ready to explain the value you add. Our guide on handling that conversation gives scripts and options.

The bottom line

Nate B Jones makes a sensible, evidence-backed case: the newest, priciest AI model is impressive, but for a growing share of everyday work it is not necessary. Businesses that test their own tasks, count the full cost and route only hard work to premium models get more from AI without a bigger bill. And if you sell services, prepare for clients who have worked that out too.

author avatar
Garry Knight
I'm Garry Knight, the person behind Prodify Digital. I write about email list building, email marketing, SEO, AI search and the tools that connect them. My aim is to make online marketing easier to understand, so creators and small business owners can make informed decisions about building an audience and keeping people engaged. Here you'll find straightforward guides and product reviews that explain what something does, where it fits and which limitations matter. The focus is on clear explanations and useful next steps—not hype, shortcuts or promises of easy earnings.
One thought on “Do You Need the Most Expensive AI Model? Why Cheaper Is Often Enough”

Leave a Reply