UK Fintech Opetek Claims 90% Reduction in AI Token Costs for Capital Markets
The company's ARIUS platform aims to curb soaring enterprise AI expenditures by processing data before it reaches large language models.
UK-based fintech Opetek claims it can slash AI token expenditures for financial institutions by as much as 90%. The company is targeting the high cost of large language model (LLM) inference in capital markets, where data volume and reasoning complexity often lead to prohibitive operational expenses.
To achieve these reductions, Opetek utilizes its core product, ARIUS, a reasoning platform designed specifically for the complexities of financial services. In a capital markets task test, Opetek reported that its system reduced LLM inference costs from approximately $0.802 to $0.07 per query, representing a 91.7% drop in cost per interaction. The ARIUS system works by integrating fragmented workflows—including quantitative market data, analytical functions, qualitative news, and chat logs—into a single reasoning model.
The Cost of AI Adoption
Financial institutions are currently grappling with soaring AI costs as adoption scales across the enterprise. This trend is often linked to the Jevons Paradox, where increases in efficiency lead to an overall increase in total consumption. In the high-stakes environment of financial services, where transaction speed and precision are paramount, the financial burden is significant. Some institutions have reported token costs reaching tens of thousands of dollars per month for individual users.
Shifting from Brute Force to Reasoning
Opetek’s approach is based on the premise that while raw computation is relatively inexpensive, the actual reasoning performed by an LLM is costly. According to Varqa Abyaneh, Founder and CEO of Opetek, firms should process data first rather than feeding entire datasets into models because of this cost disparity.
By utilizing a "trusted reasoning system" that filters and processes information before it is fed into the LLM, Opetek ensures that only the strictly necessary data is sent for reasoning. This shift away from "brute force" data feeding provides a potential blueprint for enterprises to mitigate the financial strain and "cognitive debt" associated with at-scale AI deployment. Furthermore, this method may improve the traceability and governance of AI-assisted financial decisions by narrowing the data window used for each specific output.
Future Outlook
As financial firms continue to integrate generative AI into their core operations, the scalability of Opetek's thesis will be critical. The industry is watching to see if these cost reductions can be maintained across diverse, real-world financial workflows beyond controlled test cases. While the initial figures suggest a significant path toward sustainable AI spending, the long-term impact on enterprise margins will depend on the widespread adoption of pre-processing reasoning layers over direct LLM integration.