Gemini 3.1 Pro is surprisingly good at classifying banking transactions (butternut.click)

In a recent evaluation, Gemini 3.1 Pro outperformed its peers in classifying banking transactions from a user’s South African transaction history, demonstrating remarkable proficiency and adaptability. When prompted to categorize various transactions, Gemini achieved an impressive score of 59 out of 60, while its competitors, GPT 5.2 Thinking and Claude Opus 4.6, scored 52 and 51, respectively. Notably, Gemini excelled at deciphering obscure transaction identifiers and country-specific services that posed challenges for the other models. This performance highlights Gemini’s advanced understanding of regional nuances and its ability to contextualize modern banking terms, an aspect critical for financial applications. For instance, it accurately identified "AE" as Astron Energy, reflecting its training on relevant data and contextual knowledge beyond what competitors could initially provide. Such capabilities could greatly enhance user experiences in personal finance management, making AI more accessible and effective for individual financial tracking and decision-making. This development is significant for the AI/ML community, marking a leap in the application of language models in real-world scenarios where precision and contextual understanding are key.