What happened with prices
The most recent, concrete example: one of the leading providers cut its model pricing to as low as $0.20 per million input tokens, an adjustment of up to 80% compared to previous prices. This isn't an isolated case — it's the continuation of a sustained price-drop trend across the industry, driven by competition among providers and efficiency improvements in the models themselves.
At the same time, mass adoption (some of these assistants now exceed a billion weekly users) gives providers the scale needed to keep cutting prices without losing margin.
What use cases that weren't cost-effective before now are
When the cost per token drops this sharply, it changes the business case for projects that used to get shelved:
- Processing large volumes of documents or tickets that used to be too expensive to analyse one by one (contracts, invoices, support emails).
- 24/7 conversational agents for small SMEs — previously only economically viable for companies with large budgets.
- Sentiment analysis and mass classification of customer feedback, reviews or surveys.
- Content generation at scale for marketing, while keeping in mind the labelling obligations of the AI Act that are already in force.
The concrete data point: a model that cost a certain amount to run at scale a year ago can today cost a fraction of that. A pilot project that used to require a large-company budget can now fit within an SME budget.
Mistakes to avoid now that AI is cheaper
- Overpaying for a "premium" model when a cheaper one solves the specific use case just as well.
- Scaling without measuring. Cheap doesn't mean free: high volumes still add up, and without spend control a project can blow its budget.
- Not optimising usage. Unnecessarily long prompts or lack of caching still waste money, even if the price per token has dropped.
- Choosing a model by brand instead of comparing real benchmarks for your company's specific use case.
How to take advantage of the price drop: 3 questions
- What AI project did you shelve 6-12 months ago because of cost? It's worth revisiting with today's prices.
- Are you using the most expensive model available without having compared cheaper alternatives for your specific use case?
- Do you have a narrow pilot you could launch now with a budget that didn't stretch far enough before?
Conclusion
The price drop doesn't solve every problem in an AI project — data governance mistakes, poorly defined scope, or the lack of an upfront diagnostic remain the main cause of failure. But it does remove the cost excuse for many SMEs that had been priced out until now.
At Dataverse Solutions we help choose the right model and architecture for each use case, optimising cost without sacrificing results.