The DMind Benchmark, developed by the Minara Team, has been officially accepted for inclusion in KDD 2026, becoming the first digital asset × AI large language model evaluation benchmark to be featured at a top international conference.

The DMind Benchmark, launched by the Minara team in 2025, has recently had its paper accepted by the Datasets & Benchmarks Track of the 2026 KDD main conference. It becomes the first large language model (LLM) evaluation benchmark at the intersection of digital assets and AI to pass peer review and secure a spot in the main track of an international top conference (not a workshop or demo). KDD is a leading international conference in data mining and machine learning, with an acceptance rate of roughly 29% for this track this year. The DMind Benchmark assesses models’ multi-step reasoning capabilities in real-world digital asset scenarios, comprising 3,154 objective questions and 389 open tasks. All content was reviewed by domain experts on a question-by-question basis, covering 9 subfields including DeFi, tokenomics, and contract security. It has systematically evaluated 31 leading LLMs such as GPT, Claude, and Gemini. Its dataset once topped Hugging Face’s overall Trending list, amassing over 13,000 cumulative downloads. A July 2026 retest revealed that the performance gains of six latest flagship models on the benchmark failed to bridge the domain depth gap. Leveraging this benchmark and DMind, the initiative is building a dedicated intelligent infrastructure layer for the digital asset sector, enabling AI to truly comprehend the digital asset ecosystem.