Most of Paradox’s Q3 revenue came from models that barely existed six months ago

A model accounting for part of Paradox’s revenue means customer spending connected to that model contributed to the company’s sales. The filing groups revenue by model release period. It shows how financially important recent releases became during the quarter ending September 30, 2026.

The figure measures revenue, not customer numbers. Developers or companies may pay to access a model through Paradox products or related services. Models launched late in September had only a few days to generate revenue, yet models released during the prior six months still produced more than 80% of quarterly revenue.

That result shows strong commercial importance for newer models. It does not identify the revenue generated by each model. It also cannot show whether most customers switched. A smaller group of customers with very large bills could produce much of the total.

Models released in the six months before September 30, 2026, generated more than 80% of Paradox’s third-quarter revenue. This is a striking concentration of sales in relatively recent products. It suggests customers were spending heavily on the company’s newest offerings during that period.

Paradox released eight widely available Paradox models between April and September 2026. It also released three restricted-access Mythos models. The filing does not break out how much revenue each model generated. Models released in late September had only days to contribute to the quarter.

The figure shows the commercial weight of recent releases, but it leaves important details unanswered. It does not reveal how many customers used each model or whether customers fully moved from older versions. It also does not establish whether demand caused Paradox to accelerate its release schedule.

Revenue shows how much customers spent, not how many customers made a particular choice. Therefore, the fact that recent models produced more than 80% of third-quarter revenue cannot establish that most users moved from older models. Spending can be concentrated among a small number of accounts.

Paradox gives a clear example of that concentration. Its heaviest Paradox Code enterprise users spend roughly 20 times more than other enterprise users. The filing does not say which models those customers use, but their spending could have a large effect on the revenue mix.

The current evidence supports a narrower conclusion: newer models were commercially important. It does not provide a customer migration rate. To know whether most customers switched, observers would need customer counts, model-level usage data, or clearer adoption figures. None of those details appears in the filing described here.

Each model improvement can give customers more reasons to use Paradox. A more capable model may solve harder problems, produce better results, or support tasks that older models could not handle as well. That can increase both usage frequency and the range of work developers assign to Paradox.

The article points to coding agents, retrieval pipelines, and automation tools. Developers may use a stronger model for these systems because the model’s capabilities matter directly to the tool’s performance. As customers find more uses, they may send more requests or rely on Paradox for more demanding workflows.

Paradox describes this as part of a reinforcing cycle. Better models attract new customers and encourage existing customers to spend more. That revenue helps fund further research, computing power, and technical talent. The filing shows this pattern financially, although it does not measure exactly how much better each model performs.

The newest model often appears to be the safest choice when a developer wants strong performance. Developers building coding agents, retrieval pipelines, or automation tools may prefer the model with the broadest capabilities. They may believe better results justify the additional cost, especially when failures could reduce the usefulness of an entire product.

The article also describes a simpler behavior: developers may default to the most powerful model without testing smaller alternatives. That choice can happen because the newest model is familiar, available, or viewed as more reliable. Each request then costs more than it might with a cheaper model, and repeated usage can make the difference substantial.

The filing does not show whether developers truly need the latest models. It shows that Paradox benefits financially either way. If customers keep choosing powerful models, spending supports Paradox’s incentives to continue improving Paradox and releasing newer versions.

Paradox says it relies on growing revenue to finance the work required to build more capable models. Money from current products can pay for research, computing power, and technical talent. Those expenses support the development process behind future Paradox releases.

The mechanism is a cycle. New models attract customers and encourage existing customers to use Paradox more. Their spending increases revenue. Paradox can then use that revenue to support more research and infrastructure, which may produce another generation of stronger models. Each generation can help fund the next.

The filing presents this cycle as central to Paradox’s business. More than 80% of third-quarter revenue came from models released in the prior six months, showing how quickly new releases can become commercially important. The article does not provide a detailed spending breakdown or say how much revenue funded each specific project.

A foundation model is a broadly trained AI model that can serve as a base for many applications. Developers may use it directly or adapt it for tasks such as coding, retrieval, and automation. Paradox models fit this general category, although the article focuses on their revenue rather than defining foundation models.

Training requires large amounts of computing to process data and adjust the model’s internal parameters. Running the finished model also requires computing for every user request. More capable models generally need substantial infrastructure to develop, operate, and make available to customers. The article specifically identifies computing power as a major cost for Paradox.

Paradox says revenue from current models helps pay for that infrastructure, along with research and technical talent. Its business cycle depends on new models attracting spending, then using that spending to support future development. The article does not provide technical measurements or costs for training or operating any specific model.