Integrating artificial intelligence into heavy industrial manufacturing requires specialized domain models capable of navigating strict engineering constraints, complex material variables, and large-scale operational workflows. The release of the Chemical Engineering Large Language Model (ChemELLM) 3.0 Pro by the Dalian Institute of Chemical Physics (DICP) under the Chinese Academy of Sciences—jointly developed with iFlytek, Alibaba Cloud Computing, and other partners—marks a major technological evolution. Transitioning from a basic knowledge-acquisition assistant into an active execution partner, the upgraded model provides advanced support for managing complex chemical research, engineering design, and plant operations.
The core innovation of ChemELLM 3.0 Pro lies in its sophisticated four-tier architecture, which comprises a foundational cognitive large model, intelligent execution agents, specialized professional skills and tools, and targeted industrial application scenarios. While the cognitive core handles advanced multimodal information parsing and task planning, the intelligent agents act as execution entities that independently manage process decomposition, tool invocation, result verification, and dynamic feedback loops. Evaluated against rigorous chemical-domain benchmarks, ChemELLM 3.0 Pro achieved text-based Q&A accuracy of 81.96% and multimodal Q&A accuracy of 80.75%, representing respective performance jumps of 20.2% and 31.4% over previous iterations. With cumulative API calls exceeding 14 million across more than 300 registered enterprises and research institutes, the platform has proven its commercial and scientific reliability.
Looking toward future industrial automation roadmaps, DICP plans to develop ChemELLM 4.0 to further enhance multi-tool collaborative planning and multi-step reasoning capabilities. As highlighted in coverage by People's Daily, this upcoming version aims to leverage mature industrial processes like methanol-to-olefins as validation scenarios, unifying laboratory research, engineering design, and plant operations into a seamless "lab-to-plant" transition. By reducing experimental cycle times by up to 30% and lowering R&D engineering costs, domain-specific AI platforms are accelerating the digital transformation of the global chemical manufacturing sector.
News source: https://peoplesdaily.pdnews.cn/china/er/30053058925