---
title: "Training models in-house: the tradeoff"
description: "Renting a frontier model is cheaper and faster on day one. What changes at year three, and the three conditions that make training your own worth it."
canonical: https://structime.com/en/blog/research/training-in-house/
language: en
last_modified: 2026-10-21
publisher: "AEON Systems (SIREN 993102821)"
source_format: html
---

# Training your own models: what it buys and what it costs

> Renting a frontier model is cheaper and faster on day one. What changes at year three, and the three conditions that make training your own worth it.

Source : https://structime.com/en/blog/research/training-in-house/
Publisher: AEON Systems (SIREN 993102821) · StrucTime

Calling an API is cheaper on day one, and that comparison is the one everyone runs. The interesting comparison is the one at year three.

Published 24 August 2026

Calling somebody’s API is cheaper on day one, faster to ship, and better at almost everything. That comparison is easy to run and it is the one most teams run. The comparison worth having is the one at year three.

## What renting commits you to

A price you do not set. Per-token pricing has moved in both directions and will move again, and a product whose margin depends on it inherits that volatility.

A model you do not keep. Versions are deprecated. Behaviour changes under a stable name. A prompt tuned over months against one version is not portable to the next, and the work of re-tuning is invisible until it is urgent.

A data path you have to explain. Every document sent for processing leaves the company. There are contractual answers to this, and they are answers, but they are somebody else’s promise rather than your architecture.

Renting can still be the right call. This is the list of what you trade away, and it is worth having written down before the decision rather than after.

## The three conditions

Training your own is worth it when all three hold, and it is a bad idea when any one fails.

**The task is narrow enough to be learned.** Five defined operations, not general assistance. See [small models on closed tasks](https://structime.com/en/blog/research/small-models-closed-tasks/).

**You have data nobody else has.** If your training set is public data, you are reproducing at great cost what someone else already did better. The case rests on proprietary material.

**The work outlives the hype cycle.** Training is a capability you build over years. If the requirement might disappear in eighteen months, rent.

## What it costs

Hardware you own and have to keep busy, which is a different discipline from paying per call. Engineers who understand training rather than prompting. And an evaluation practice, because a model you trained yourself comes with no external benchmark and no vendor claiming it works. You have to know that it does, which means [baselines](https://structime.com/en/blog/research/baselines-that-matter/) and a held-out set of real documents.

## Where we stand

We train our own models in France, on our own machines. Our language models and our world model are work in progress, and nothing has been published so far. That status is on the [research page](https://structime.com/en/research/) and it is deliberately unflattering, because the alternative is a claim we would have to walk back.

A country that controls neither its software nor its artificial intelligence ends up depending on those who do. So we build them ourselves. That is the reason underneath the arithmetic, and it does not replace the arithmetic.

How we train and evaluate our own models, what we can claim today, and what we deliberately do not claim yet.

## More on this topic

- [What a world model is, and what it is not The term covers work as different as video generation and industrial forecasting. Stripped of marketing, it means one thing: a learned representation of how a system changes.](https://structime.com/en/blog/research/what-is-a-world-model/)
- [The baselines that decide whether a forecast is any good A model that predicts an industrial signal with two percent error sounds strong until you learn that repeating yesterday's value gets one and a half.](https://structime.com/en/blog/research/baselines-that-matter/)
- [Small language models on closed tasks: the case for narrow Most business uses of a language model repeat the same five operations on documents that follow a known shape.](https://structime.com/en/blog/research/small-models-closed-tasks/)