--- title: "Is one AI model enough? When to use more than one | Len P. van der Hof" description: "One AI model is fine for low-volume, single-task work with no sensitive data. Past that, you need a portfolio: written choices, a fallback, and a review date." image: "https://lenvanderhof.com/media/generated/blog-hero-one-model-is-not-a-portfolio-v2.e1368364de82.wide.webp" --- [AI Systems](https://lenvanderhof.com/en/blog/category/ai-systems/) Research guide # Is one AI model enough? When a single model works, and when you need a portfolio The number of models is the wrong question. The right one: did anyone write down why each model runs, and what happens when it stops? Len P. van der HofPublished 29 September 20267 min read One pot for every dish is a habit, not a kitchen. Direct answer One AI model is enough when volume is low, there is one kind of task, and no regulated or confidential data is involved. Outside that case, one model for everything means paying top rates for easy work and having no plan when that model fails or is retired. A model portfolio is the fix: a written list of which model does which job, why, what takes over when it fails, and when the choice is reviewed. More models without those rules is just a collection. ## Key takeaways - One model is fine for low volume, one task type, and no regulated data. - Past that, one model is too expensive for the easy work or too weak for the hard work. - Providers retire models on their own schedule. A model with no replacement plan is a deadline you did not choose. - More models is not a portfolio. Twelve models from one provider still share one point of failure. - A portfolio is written down: which model, which job, why, what takes over, and when it is reviewed. **One AI model is enough when the work is small, uniform, and safe to send out: low volume, one kind of task, and no regulated or confidential data.** Past that point, the useful question is not how many models you run. It is whether each choice is written down, owned, and reviewed. That written set of choices is a model portfolio. A pile of models without it is a collection. One model can be a decision. One model because nobody ever decided is a risk that grows every month. ## When is one model enough? [The Model Portfolio](https://lenvanderhof.com/books/the-model-portfolio/), my book on running several AI models, makes the concession plainly: with low volume, one task type, and no regulated data, the overhead of managing several models may genuinely not pay. Use this table to see which side you are on. One model is fine when…You need a portfolio when…Volume is low and the bill is smallVolume grows and the bill starts to matterThere is one kind of taskTasks differ in difficulty, speed, or formatNo regulated or confidential data is involvedSome data may not leave your controlA day without the model is an inconvenienceAn outage or a retired model would stop the work Most products start in the left column. Few stay there, because every new feature reaches for the endpoint (the address your software calls to reach a model) that already exists. ## What goes wrong with one model for everything? Three things, and they arrive slowly enough that nobody schedules a decision. **You pay the wrong price somewhere.** A single model chosen for your hardest task is expensive for your easiest ones. A single model chosen for cost is too weak for your hardest ones. The book tells a composite story (not a real company) that shows how it happens. A product launched with one task and one frontier model, the most capable and most expensive tier. By the time anyone looked, seven jobs ran through that one endpoint: autocomplete, nightly ticket summaries, entity extraction (pulling names, dates, and amounts out of text), an FAQ bot, document parsing, notification drafts, and a weekly research synthesis. When someone finally mapped what each job needed, four could wait for a batch run and needed little reasoning, and two wanted speed more than depth. Six of the seven were candidates for a cheaper workhorse model, pending tests. Nobody had chosen that arrangement. It had simply grown. **The model can disappear.** Providers retire models on their own schedule. Anthropic’s [model deprecations page](https://platform.claude.com/docs/en/about-claude/model-deprecations), for example, promises at least 60 days’ notice before a publicly released model is retired, and states that requests to a retired model will fail. On 5 June 2026 it notified developers that Claude Opus 4.1 would retire on 5 August 2026. If that was your only model, you had a migration deadline you did not pick. The migration is also not a one-line change. The book treats prompts as assets tied to one model: a prompt tuned for one model can perform worse on another, so every switch needs testing. **Some data may not go there at all.** If part of your work involves data that must stay on your own infrastructure or in one region, a single hosted model cannot serve it. In the book’s terms, privacy is a hard routing constraint, not a preference. It overrides cost and capability. ## Why is “more models” not the answer either? Adding models without rules produces the opposite problem. The book describes it: six months later the stack has five models, nobody is sure who approved three of them, and the monthly bill explains nothing. More models also do not automatically mean less risk. The book makes the point with a list of twelve models from one provider: it has diversified nothing at the provider level, because a change of terms, a long outage, or an acquisition hits all twelve at once. A fallback (the path that runs when the primary model fails) from the same provider protects you against a broken model, not against a broken provider. So the line between a collection and a portfolio is not the count. It is whether each model has a written job, an owner, a budget, the data it may touch, a fallback, and a review date. ## What turns a set of models into a portfolio? The book’s answer is ROUTE, a five-step discipline for running a mix of AI models on purpose. The book is available in English and Dutch. You do not need it to use the steps. - **Register.** Keep one list of every model you run, with its job, owner, cost class, the data it may see, and its retirement status. A model that is not on the list runs unowned. - **Objective typing.** Before you choose a model, write down what the task needs: reasoning depth, output format, speed, quality bar, and privacy. - **Utilize policy.** Write the rule that sends each task to a model, including the fallback and the point where the work stops rather than going somewhere it should not. - **Track.** Measure cost, speed, and quality per task, not only the monthly total, so you can see which choice spends the money. - **Evolve.** Promote, demote, and retire models on a schedule. When an index such as [Undominated.ai](https://lenvanderhof.com/undominated/) shows a model that scores at least as high for less, treat it as a reason to test, not a reason to switch. The routing rule itself has its own page: [What is model routing?](https://lenvanderhof.com/en/blog/what-is-model-routing/). Two patterns that rule can use are a [model cascade](https://lenvanderhof.com/en/blog/what-is-a-model-cascade/) (try the cheap model first, escalate on a written trigger) and the split between [frontier, workhorse, and local](https://lenvanderhof.com/en/blog/frontier-vs-workhorse-vs-local/) models. ## What does one written choice look like? Here is a route card for one task. The company and the numbers are hypothetical. Copy the left column and fill in your own busiest task. FieldExample: nightly support-ticket summariesTaskSummarise yesterday’s closed tickets for the product teamNeedsDone by 07:00; 150 words per ticket; little reasoning; contains customer names, EU processing onlyPrimary modelA workhorse model on an EU-hosted endpointWhy this oneCheapest registered model that passed a 50-ticket testFallbackA workhorse model from a different provider, same EU data terms, tested on the same 50 ticketsStop ruleIf no EU-eligible model is available, hold the batch. Never send the tickets elsewhere.OwnerOne named personTrackCost per summary, share of summaries the team rejects, finish timeReviewEvery quarter, or as soon as a provider announces a retirement The card follows ROUTE. The models come from your Register (hence “registered model” in the “Why this one” row), the “Needs” row is Objective typing, the model, fallback, and stop rule are Utilize policy, the “Track” row is Track, and the review date is Evolve. Notice what the card does not say: how many models you need. It might turn out that one model covers every task you have. Then you have a portfolio of one, and you know why. ## Try this today (15 minutes) This is the field test from the opening chapter of the book. 1. Open your AI provider’s billing or usage dashboard. 2. Find the model that used the largest share of tokens (the chunks of text you are billed for) last month. 3. Write down what share of those requests genuinely needed that model’s full ability. 4. If you cannot answer, you do not have a routing strategy yet. You have a default. Fill in one route card for your busiest task. The card takes fifteen minutes. The next retirement notice will not wait that politely. Cite this:Is one AI model enough? When a single model works, and when you need a portfolio.Len P. van der Hof. [https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/](https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/) · Published 29 September 2026. ## Terminology - [ROUTE](https://lenvanderhof.com/glossary/route/) ## Sources 1. [Model deprecations](https://platform.claude.com/docs/en/about-claude/model-deprecations) · Anthropic 2. [ROUTE](https://lenvanderhof.com/frameworks/route/) 3. [The Model Portfolio](https://lenvanderhof.com/books/the-model-portfolio/) 4. [What is model routing?](https://lenvanderhof.com/en/blog/what-is-model-routing/) 5. [Undominated.ai](https://lenvanderhof.com/undominated/) ## Further reading - [ROUTE](https://lenvanderhof.com/frameworks/route/) - [The Model Portfolio](https://lenvanderhof.com/books/the-model-portfolio/) - [What is model routing?](https://lenvanderhof.com/en/blog/what-is-model-routing/) - [What is a model cascade?](https://lenvanderhof.com/en/blog/what-is-a-model-cascade/) - [Frontier, workhorse, or local](https://lenvanderhof.com/en/blog/frontier-vs-workhorse-vs-local/) About the author ## [Len P. van der Hof](https://lenvanderhof.com/en/authors/len-p-van-der-hof/) Entrepreneur, AI Innovator and Venture Builder Len P. van der Hof builds practical AI systems, digital ventures and evidence-informed tools for founders. ```json { "@context": "https://schema.org", "@graph": [ { "@type": "Person", "@id": "https://lenvanderhof.com/#person", "name": "Len P. van der Hof", "alternateName": [ "Len van der Hof", "L.P. van der Hof", "Leendert Pieter van der Hof" ], "honorificSuffix": "MSc", "url": "https://lenvanderhof.com/", "image": [ "https://lenvanderhof.com/photos/len-portrait-1.jpg", "https://lenvanderhof.com/photos/len-portrait-2.jpg", "https://lenvanderhof.com/photos/len-portrait-3.jpg", "https://lenvanderhof.com/photos/len-portrait-4.jpg", "https://lenvanderhof.com/photos/len-speaking.jpg", "https://lenvanderhof.com/photos/len-hero.jpg" ], "jobTitle": "Entrepreneur, AI Innovator and Venture Builder", "description": "Len P. van der Hof, MSc, is a Dutch entrepreneur and AI innovator in Zwijndrecht. He builds ReasonKit, MindSesh, Undominated.ai, books under his name, the fiction imprint LPH98.lifestyle, and technology ventures through LPH98.ventures. Eleven titles in Systems for the Strategic Self are available now, in English and Dutch.", "address": { "@type": "PostalAddress", "addressLocality": "Zwijndrecht", "addressCountry": "NL" }, "alumniOf": { "@type": "CollegeOrUniversity", "name": "Rotterdam School of Management, Erasmus University" }, "knowsAbout": [ "Artificial intelligence", "AI agents", "Agentic AI systems", "LLM routing", "SEO", "Generative engine optimization", "Answer engine optimization", "Venture building", "Founder performance", "Founder psychology", "Evidence-based decision-making" ], "sameAs": [ "https://www.linkedin.com/in/lenvanderhof/", "https://x.com/LenvanderHof", "https://www.youtube.com/channel/UCTG20buKqYYbitqqf7l3zJA", "https://www.instagram.com/Lenvanderhof/", "https://www.threads.com/@lenvanderhof", "https://github.com/Lenvanderhof", "https://huggingface.co/LPH98", "https://www.npmjs.com/~lenvanderhof", "https://www.goodreads.com/author/show/70983905.Len_P_van_der_Hof", "https://www.amazon.com/author/lenvanderhof", "https://www.bol.com/nl/nl/b/len-p-van-der-hof-msc/609879394/", "https://bsky.app/profile/lenvanderhof.com", "https://mastodon.social/@Lenvanderhof", "https://crates.io/users/Lenvanderhof", "https://cursor.com/@Lenvanderhof", "https://medium.com/@Lenvanderhof", "https://gitlab.com/Lenvanderhof", "https://hub.docker.com/u/lenvanderhof/", "https://dev.to/lenvanderhof", "https://www.facebook.com/Lenvanderhof", "https://soundcloud.com/Lenvanderhof" ], "affiliation": [ { "@id": "https://lenvanderhof.com/#publisher" }, { "@id": "https://lenvanderhof.com/#mindsesh" }, { "@id": "https://lenvanderhof.com/#lifestyle" } ] }, { "@type": "WebSite", "@id": "https://lenvanderhof.com/#website", "url": "https://lenvanderhof.com/", "name": "Len P. van der Hof", "description": "Len P. van der Hof, MSc, is a Dutch entrepreneur and AI innovator in Zwijndrecht. He builds ReasonKit, MindSesh, Undominated.ai, books under his name, the fiction imprint LPH98.lifestyle, and technology ventures through LPH98.ventures. Eleven titles in Systems for the Strategic Self are available now, in English and Dutch.", "inLanguage": [ "en", "nl" ], "publisher": { "@id": "https://lenvanderhof.com/#person" } }, { "@type": "Organization", "@id": "https://lenvanderhof.com/#publisher", "name": "LPH98.ventures", "url": "https://lph98.ventures", "founder": { "@id": "https://lenvanderhof.com/#person" } }, { "@type": "Organization", "@id": "https://lenvanderhof.com/#mindsesh", "name": "MindSesh", "url": "https://mindsesh.net", "founder": { "@id": "https://lenvanderhof.com/#person" } }, { "@type": "SoftwareApplication", "@id": "https://lenvanderhof.com/#reasonkit", "name": "ReasonKit", "url": "https://reasonkit.sh", "creator": { "@id": "https://lenvanderhof.com/#person" } }, { "@type": "SoftwareApplication", "@id": "https://lenvanderhof.com/#undominated", "name": "Undominated.ai", "url": "https://undominated.ai", "creator": { "@id": "https://lenvanderhof.com/#person" } }, { "@type": "Organization", "@id": "https://lenvanderhof.com/#lifestyle", "name": "LPH98.lifestyle", "url": "https://lph98.lifestyle", "founder": { "@id": "https://lenvanderhof.com/#person" } }, { "@type": "ImageObject", "@id": "https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/#primaryimage", "url": "https://lenvanderhof.com/media/generated/blog-hero-one-model-is-not-a-portfolio-v2.e1368364de82.wide.webp", "contentUrl": "https://lenvanderhof.com/media/generated/blog-hero-one-model-is-not-a-portfolio-v2.e1368364de82.wide.webp", "representativeOfPage": true }, { "@type": "BreadcrumbList", "@id": "https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/#breadcrumb", "itemListElement": [ { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://lenvanderhof.com/" }, { "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://lenvanderhof.com/en/blog/" }, { "@type": "ListItem", "position": 3, "name": "AI Systems", "item": "https://lenvanderhof.com/en/blog/category/ai-systems/" }, { "@type": "ListItem", "position": 4, "name": "Is one AI model enough? When a single model works, and when you need a portfolio", "item": "https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/" } ] }, { "@type": "WebPage", "@id": "https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/#webpage", "url": "https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/", "name": "Is one AI model enough? When a single model works, and when you need a portfolio", "description": "One AI model is fine for low-volume, single-task work with no sensitive data. Past that, you need a portfolio: written choices, a fallback, and a review date.", "isPartOf": { "@id": "https://lenvanderhof.com/#website" }, "primaryImageOfPage": { "@id": "https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/#primaryimage" }, "breadcrumb": { "@id": "https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/#breadcrumb" }, "inLanguage": "en-GB" }, { "@type": "BlogPosting", "@id": "https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/#article", "mainEntityOfPage": { "@id": "https://lenvanderhof.com/en/blog/one-model-is-not-a-portfolio/#webpage" }, "headline": "Is one AI model enough? When a single model works, and when you need a portfolio", "description": "One AI model is fine for low-volume, single-task work with no sensitive data. Past that, you need a portfolio: written choices, a fallback, and a review date.", "datePublished": "2026-09-29T14:00:00.000Z", "author": { "@id": "https://lenvanderhof.com/#person" }, "publisher": { "@id": "https://lenvanderhof.com/#person" }, "image": [ "https://lenvanderhof.com/media/generated/blog-hero-one-model-is-not-a-portfolio-v2.e1368364de82.square.webp", "https://lenvanderhof.com/media/generated/blog-hero-one-model-is-not-a-portfolio-v2.e1368364de82.landscape.webp", "https://lenvanderhof.com/media/generated/blog-hero-one-model-is-not-a-portfolio-v2.e1368364de82.wide.webp" ], "articleSection": "AI Systems", "inLanguage": "en-GB" } ] } ```