British businesses relying on Anthropic’s Claude family of large language models may be paying more than they realise, due to the company’s unusually extravagant tokeniser. Tokenisers break text into smaller units — tokens — which models process and price per unit. Anthropic’s tokeniser uses far more tokens than rivals for the same input, effectively inflating costs without clear disclosure.
For a typical 1,000-word business document, Anthropic’s tokeniser might consume 30–40 per cent more tokens than OpenAI’s GPT-4 or Google’s Gemini. This discrepancy is not immediately obvious to developers, who see per-token pricing but not the token count per request. UK startups and enterprises deploying Claude for customer support, document analysis or code generation could face bills significantly higher than anticipated.
The issue has regulatory implications. The UK Information Commissioner’s Office (ICO) has called for greater transparency in AI pricing and performance metrics. Meanwhile, the EU AI Act, which applies to UK firms serving European customers, requires providers to disclose the computational resources used by their models. Tokenisation efficiency falls squarely into that category, yet Anthropic has not publicly detailed its tokeniser’s design or efficiency benchmarks.
Dr. Helena Cross, a lecturer in AI economics at the University of Cambridge, said: “Tokenisation is not just a technical curiosity — it directly affects the bottom line. If one provider’s tokeniser is significantly less efficient, businesses are effectively subsidising that inefficiency. Without standardised reporting, it’s impossible for procurement teams to make like-for-like comparisons.” She added that the lack of transparency could erode trust in AI pricing models, particularly for cost-sensitive small and medium-sized enterprises.
For UK consumers, the impact is indirect but real. Businesses that pass on higher AI costs may raise prices for AI-enhanced products and services. In sectors like legal tech, fintech and healthcare, where AI is increasingly embedded, hidden token inflation could widen the gap between large enterprises that can absorb costs and smaller players that cannot. The broader UK economy risks slower AI adoption if pricing remains opaque and unpredictable.