Blind test of seven frontier models reveals unexpected budget performance

A recent blind comparison conducted by researcher Nufar Gaspar and featured on the AI Daily Brief examined how seven leading artificial intelligence models performed across six everyday practical assignments. Rather than relying on official benchmark tables published by tech companies, the trial tested options ranging from high end systems like Opus 5.5 to lower cost alternatives like GPT 6.1 Sol and Kimi K3 without revealing their identities during evaluation.

The results showed that expensive frontier systems are not always necessary for routine work. While premium options like Opus 5.5 excelled at complex creative tasks such as building a course website, budget friendly options matched top tier quality on standard writing and analysis. For instance, GPT 6.1 Sol achieved human ratings equal to higher priced rivals at a fraction of the price, while Kimi K3 offered the fastest response times and lowest operational costs. In contrast, Grok 4.7 proved to be significantly slower than its competitors.

The trial also highlighted how difficult automated evaluation remains, as an automated AI judge agreed with human scoring on only one of the six tasks. This disagreement occurred largely because the software judge analyzed raw code instead of visual layout during design tests. These findings suggest that corporate buyers could save substantial money by delegating everyday administrative work to cheaper models rather than buying expensive licenses for every employee. However, as new models update every few weeks, it remains unclear whether companies can maintain such custom testing routines without wasting valuable staff hours.