Norwegian firm says companies waste tokens on AI queries

Norwegian company Rock Sovereign says many businesses are spending far more on AI than they need to, and that cutting the bill is simple. Its managing director, Paal Hatledal, claims companies can reduce their AI costs by 40 to 50 percent.

The problem lies in how companies use external language models to answer questions about their own documents. Relevant pieces of text have to be sent along with every single request, and the more text that is sent, the higher the cost, since pricing is based on tokens. Rock Sovereign, which was recently founded, argues that many organisations send much more than the model actually needs.

Its solution is to split content into small, natural paragraphs at the point when the data is indexed. That way only the most relevant pieces are passed on to the model, rather than large chunks of text.

The software behind the approach comes from Tellusr. Its managing director, Morten Krogh-Moe, describes the difference this way: instead of sending big chunks to large models, which use more tokens, the system sends smaller and more precise ones.

The piece, published on 12 November 2026, presents the 40 to 50 percent saving as the company's own claim, and no independent figures or customer results are given. It matters for any organisation running document-based question answering on external models, where token use drives the running cost.