AI token costs: Opetek claims 90% cut for enterprises

What happened

UK fintech Opetek has put forward a method it says can cut enterprise spending on AI "tokens" — the metered units that large language model providers charge for processing input and output — by as much as 90%. The claim, reported by diginomica, comes as organisations scaling up generative AI deployments increasingly find themselves exposed to token-based pricing that rises in step with usage.

Opetek's pitch centres on the idea of "tokenomics": treating token consumption as a cost and risk variable that needs active management, rather than an unavoidable by-product of running AI models. The company argues that as businesses move from pilot projects to production-scale AI, controlling this cost becomes inseparable from the parallel challenge of ensuring AI outputs are accurate and trustworthy enough to base real decisions on.

Why it matters

Token-based pricing has quietly become one of the biggest hidden costs of enterprise AI adoption. As organisations move generative AI from experimentation into everyday workflows — customer service bots, document processing, decision support — the volume of tokens consumed multiplies quickly, and costs can scale in ways that are hard to forecast or govern. A credible route to reducing that spend by an order of magnitude would materially change the economics of deploying AI at scale, potentially making previously marginal use cases viable.

The claim also speaks to a broader tension in enterprise AI strategy: cost efficiency cannot come at the expense of output quality. Any approach that reduces token consumption will only matter commercially if it preserves — or improves — the reliability of the decisions AI systems produce, since trust in outputs is what ultimately determines whether organisations are willing to let AI touch consequential processes.

The Renascence take

Cost-per-token headlines are easy to publish and hard to verify, but the underlying instinct is sound: AI economics deserve the same scrutiny that organisations apply to any other operating cost.

Most leaders are still pricing AI the way they priced software — as a fixed cost to be budgeted once. Token-based billing behaves more like a variable cost of goods sold, and it punishes vague prompts, redundant calls and poorly scoped use cases just as surely as waste punishes a factory floor. Before chasing vendor claims of dramatic savings, customer-obsessed operators should first instrument their own AI usage: know which journeys consume the most tokens, whether that consumption maps to real customer or business value, and whether accuracy is being traded for speed anywhere in the chain. Cost discipline and decision quality are the same problem, not two separate ones.